Given a Google search amounts to 0.0003 kWh of energy consumption per search, and based on the older 3 Wh estimate, two years ago Alphabet Chairman John Hennessey said that an LLM exchange would probably cost 10 times more than a Google search in energy. If Epoch AI's new estimate is correct, it seems that a likely GPT-4o interaction actually consumes the same amount of energy as a Google search.

Server energy use isn't something that tends to cross most people's minds while using a cloud service—the 'cloud' is so far removed from our homes that it seems a little ethereal. I know I often forget there are any additional energy costs at all, other than what my own device consumes, when using ChatGPT.

Thankfully I'm not a mover or a shaker in the world of energy policy, because of course LLM interactions consume energy. Let's not forget how LLMs work: they undertake shedloads of data training (consuming shedloads of energy), then once they've been trained and are interacting, they still need to pull from gigantic models to process even simple instructions or queries. That's the nature of the beast. And that beast needs feeding energy to keep up and running.

It's just that apparently that's less energy than we might have originally thought on a per-interaction basis: "For context, 0.3 watt-hours is less than the amount of electricity that an LED lightbulb or a laptop consumes in a few minutes. And even for a heavy chat user, the energy cost of ChatGPT will be a small fraction of the overall electricity consumption of a developed-country resident."

Epoch AI explains that there are a few differences between how it's worked out this new estimate and how the original 3 Wh estimate was calculated. Essentially, the new estimate uses a "more realistic assumption for the number of output tokens in a typical chatbot usage", whereas the original estimate assumed output tokens equivalent to about 1,500 words on average (tokens are essentially units of text such as a word). The new one also assumes just 70% of peak server power and computation being performed on a newer chip (Nvidia's H100 rather than an A100).

All these changes—which seem reasonable to my eyes and ears—paint a picture of a much less power-hungry ChatGPT. However, Epoch AI points out that "there is a lot of uncertainty here around both parameter count, utilization, and other factors". Longer queries, for instance, it says could increase energy consumption "substantially to 2.5 to 40 watt-hours."

It's a complicated story, but should we expect any less? In fact, let me muddy the waters a little more for us.

We also need to consider the benefits of AI for energy consumption. A productive technology doesn't exist in a vacuum, after all. For instance, use of AI such as ChatGPT could help bring about breakthroughs in energy production that decrease energy use across the board. And use of AI could increase productivity in areas that reduce energy in other ways; for instance, a manual task that would have required you to keep your computer turned on and consuming power for 10 minutes might be done in one minute with the help of AI.

On the other hand, there's the cost of AI training to consider. But on the peculiar third hand—where did that come from?—the benefits of LLM training are starting to plateau, which means there might be less large-scale data training going forwards. Plus, aren't there always additional variables? With Google search, for instance, there's the presumed cost of constant web indexing and so on, not just the search interaction and results page generation.

In other words, it's a complicated picture, and as with all technologies, AI probably shouldn't be looked at in a vacuum. Apart from its place on the mathematician's paper, energy consumption is never an isolated variable. Ultimately, what we care about is the health and productivity of the entire system, the economy, society, and so on. As always, such debates require consideration of multi-multi-variate equations in a cost-benefit analysis, and it's difficult to get the full picture, especially when much of that picture depends on an uncertain future.

Which somewhat defines the march of capitalism, does it not? The back and forth 'but actually' that characterises these discussions gets trampled under the boots of the technology which marches ahead regardless.

And ultimately, while this new 0.3 Wh estimate is certainly a pleasant development, it's still just an estimate, and Epoch AI is very clear about this: "More transparency from OpenAI and other major AI companies would help produce a better estimate." More transparency would be nice, but I won't hold my breath.

Way back in 2017, Meta (at that time just called 'Facebook') talked a big game about “typing by brain.” Fast forward to now and Meta has shared news of two breakthroughs that make those earlier claims seem more substantial than a big sci-fi thought bubble (via MIT Technology Review). Firstly, Meta announced research that has created an AI model which "successfully decodes the production of sentences from non-invasive brain recordings, accurately decoding up to 80% of characters, and thus often reconstructing full sentences solely from brain signals."

The second study Meta shared then examines how AI can facilitate a better understanding of how our brains slot the Lego bricks of language into place. For people who have lost the ability to speak after traumatic brain injuries, or who otherwise have complex communication needs, all of this scientific research could be genuinely life-changing. Unfortunately, this is where I burst the bubble: the 'non-invasive' device Meta used to record brain signals so that they could be decoded into text is huge, costs $2 million, and makes you look a bit like Megamind.

Dated reference to an animated superhero flick for children aside, Meta has been all about brain-computer interfaces for years. More recently they've even demonstrated a welcome amount of caution when it comes to the intersection of hard and 'wet' ware.

This time, the Meta Fundamental Artificial Intelligence Research (FAIR) lab collaborated with the Basque Center on Cognition, Brain and Language, to record the brain signals of 35 healthy volunteers as they typed. Those brain signals were recorded using the aforementioned, hefty headgear—specifically a MEG scanner—and then interpreted by a purposefully trained deep neural network.

Meta wrote, "On new sentences, our AI model decodes up to 80% of the characters typed by the participants recorded with MEG, at least twice better than what can be obtained with the classic EEG system."

This essentially means that recording the magnetic fields produced by the electrical currents within the participants' brains resulted in data the AI could more accurately interpret, compared to just recording the electrical activity itself via an EEG. However, by Meta's own admission, this does not leave the research in the most practical of places.

For one, MEG scanners are far from helmets you can just pop on and off—it's specialised equipment that requires patients to sit still in a shielded room. Besides that, this study used a comparatively tiny sample size of participants, none of whom had a known traumatic brain injury or speech difficulties. This means that it's yet to be seen just how well Meta's AI model can interpret for those who really need it.

Still, as a drop out linguist myself, I'm intrigued by Meta's findings when it comes to how we string sentences together in the first place. Meta begins by explaining, "Studying the brain during speech has always proved extremely challenging for neuroscience, in part because of a simple technical problem: moving the mouth and tongue heavily corrupts neuroimaging signals." In light of this practical reality, typing instead of speaking is kind of genius.

So, what did Meta find? It's exactly like I said before: Linguistic Lego bricks, baby. Okay, that's an oversimplification, so I'll quote Meta directly once more: "Our study shows that the brain generates a sequence of representations that start from the most abstract level of representations—the meaning of a sentence—and progressively transform them into a myriad of actions, such as the actual finger movement on the keyboard [...] Our results show that the brain uses a ‘dynamic neural code’—a special neural mechanism that chains successive representations while maintaining each of them over long time periods."

To put it another way, your brain starts with vibes, unearths meaning, daisy chains those Lego bricks together, then transforms the thought into the action of typing…yeah, I would love to see the AI try to interpret the magnetic fields that led to that sentence too.

I'm in. I intone the words with a sense of victory as I navigate the file directory using only shell commands—a feat that might have impressed the occasional adolescent maybe two decades ago. But then, the camera pans out to reveal a PDF document… inside a Chrome browser… running on Windows.

Yes, this is Linux running in a PDF, running in a browser, on my Windows PC.

This completely unexpected turn, brought to you by Ading2210, the same high school student who gave you Doom running in a PDF. On YouTube, they go by vk6 (via Hackaday).

In the video description showcasing the LinuxPDF project, they explain: "I got Linux running inside a PDF file via a RISC-V emulator compiled to Javascript."

What a world we're in today: a world of such compute power that means high-level and rather ubiquitous technologies such as Javascript can run entire emulators, apparently inside PDFs. The fact that PDF documents allow Javascript to execute is a double-edged sword, of course, as while it can allow you to run DOOM and now, apparently, Linux, it can also put you at risk of dodgy malware scripts.

LinuxPDF can run in any Chromium-based browser, which includes Chrome (duh), Brave, Edge, and Opera. You can check it out for yourself here.

Of course, you're not getting the Ubuntu experience inside your Chromium-powered PDF, rather you're getting an incredibly barebones command line experience via TinyEMU RISC-V emulation.

And you're not getting a particularly fast version of that, either, thanks to the layers of emulation. You get a command line, plus a virtual keyboard to press—although you can also type your inputs using your own keyboard using the space at the bottom-right. It's a little janky (backspace only seems to register on the virtual keyboard, for instance) but what do you expect?

Ading2210 explains: "It works by using a separate text field for each row of pixels in the screen, whose contents are set to various ASCII characters." Pretty ingenious, if you ask me.

So, first DOOM, and now Linux. What's next? Crysis? How about a PDF reader running from an emulated OS running inside a PDF? We're waiting, Ading.

In an announcement tweet that has amassed well over 300,000 likes as of the time of writing, the WinRAR X account makes a pitch for why you should get its snazzy new WinRAR bag.

"What better way to support the software you’ve NEVER paid for than by buying a WinRAR bag? Do it! We dare y’all!"

The bag itself is based on the WinRAR book logo and uses the place of the magenta, indigo, and blue books as your storage. To thoroughly test how much space the 14 cm x 7 cm x 21.4 cm bag could fit, the team at Tern managed to fit 805 single-sleeved Yu-Gi-Oh cards into it. If you want to double-sleeve your cards to bring them to a tournament, you can expect to fit a little less. WinRAR later confirmed that the bag could fit three Diet Cokes. Though the original post doesn't clarify, I assume they are talking about cans.

The WinRAR bag is officially supported merch made by the aforementioned site called Tern and, as well as stocking the aforementioned bag, you can get a $237 varsity jacket with WinRAR branding. Both WinRAR products are currently being made to order due to high demand so future orders won't ship out until April.

Part of the reason this post went viral is because there's quite a lot of early internet nostalgia around WinRAR. You don't need a license to download or use WinRAR so the company has been seen as one of the good guys by sects of the internet for some time.

Even now, you can download WinRAR directly from its site without having to pay. You are encouraged to pay for WinRAR when you open the app but you can close the popout and continue to compress or extract your files. This is what the viral tweet is making a pretty tongue-in-cheek reference to.

However, with the adoption of native RAR support in Windows in 2023, WinRAR has become even more niche. This merch collaboration cashes in on the strong branding of WinRAR, and has resulted in a pretty cute bag at the same time. WinRAR has been interacting with customers showing theirs off over the last few days and the 'high demands' declaration on the website suggests the collaboration has started out strong.

My heart is definitely willing to pop this bookish bag into my cart but to be frank, my wallet might not be. A $150 bag wouldn't usually be on my radar but I can't say I'm not tempted.

The offer to purchase OpenAI is the latest twist in a dispute that goes back almost a full year: Musk sued OpenAI in March, claiming the company had abandoned "its mission to develop AGI for the benefit of humanity" in exchange for the pursuit of profit. OpenAI was founded as a non-profit but set up a for-profit subsidiary in 2019 after Musk's departure, and the money has been flowing free and fast ever since.

That all sounds noble and altruistic, but in December 2024 OpenAI released a statement saying Musk had pushed a for-profit structure for OpenAI in 2017, prior to his departure, and had demanded majority equity in and the role of CEO of the for-profit division. OpenAI ultimately rejected the offer, telling Musk it worried that "as the company makes genuine progress towards AGI, you will choose to retain your absolute control of the company despite current intent to the contrary." Musk then proposed a deal between OpenAI and Tesla in order to facilitate fundraising; shortly after that offer was rejected, Musk resigned. In 2023 he founded his own competing AI venture, xAI.

The whole thing brings to mind Musk's infamous Twitter fiasco of 2022, when the world's richest man made an off-the-cuff offer to buy Twitter for $44 billion and then did his best to get out of it before months of legal wrangling arm-twisted him into closing the deal. But the big difference in the OpenAI offer is that it's not just social media dick-swinging that's spiralled out of control. The WSJ says the proposal is backed by several big-money investors, and that means two things: It's a serious offer, and it (maybe) won't be derailed by Musk's erratic behavior.

It's hard to say how this all might work out. $97.4 billion is an awful lot of money, but a recent funding round put OpenAI's valuation at $300 billion, and the company is also leading a joint venture called Stargate that will see the US government sink $500 billion into the development of AI infrastructure. Numbers basically have no meaning at that point, except to the extent that a sub-$100 billion offer may not be enough to get the job done, particularly given Musk's handling of Twitter, now called X, which has struggled badly under his tenure.

If the deal does eventually go through, it would presumably leave Musk with a significant and possibly controlling interest in OpenAi; how that would impact OpenAI—whether it might be merged with xAI, somehow intertwined with Tesla and/or SpaceX, or just left to be mauled by DeepSeek—is anyone's guess.

Whatever the future holds, it already looks like personal beef will be a big and entertaining part of it. Altman alluded to X's financial troubles and post-Musk value plunge in his response to the offer, writing, "No thank you but we will buy Twitter for $9.74 billion if you want." In a one-word response, Musk called Altman a "swindler." He later posted a brief video clip of Altman testifying before the US senate, calling him "Scam Altman."

Point two of his "three observations" regarding the oncoming AGI future discusses the idea that "the cost to use a given level of AI falls about 10x every 12 months, and lower prices lead to much more use."

Altman goes on to say that Moore's law "changed the world at 2x every 18 months; this is unbelievably stronger."

Moore's Law is defined as the principle that the speed and capability of computers can be expected to double every two years, due to the increase in transistor counts between generations of hardware, for a minimal cost.

As a guiding principle for measuring chip development it's widely regarded to be no longer true, with Nvidia CEO Jensen Huang often noting it's dead and buried, although it's still often cited as a frame of reference for technological development as a whole.

However, AI development appears to have moved at a relentless pace in recent years, and OpenAI has been at the forefront of it with various iterations of its ChatGPT AI chatbot.

According to Altman, Artificial General Intelligence is the next step, and is already moving apace—and if his observations prove accurate, the gigantic price drops in relation to AI usage over time might be a suitable metric to define AI development in a similar way.

Altman's Law then, perhaps. Or so he likely hopes. Moore's Law was named after an observation by Gordon Moore, an Intel co-founder who observed in 1965 that the number of components in an integrated circuit was doubling every year, before adjusting his theory in 1975 to doubling every two years.

It became a guiding principle in the semiconductor industry to ensure future-resistant long term planning, although has fallen out of favor since.

The real meat and potatoes of Altman's observations surrounds what he sees as the AGI-led future to come, and how it may help unlock the creativity and productivity of us mere worker drones due to what he sees as the fabulous potential of the tech.

Still, if you happen to create a new law while musing away on your blog, that's a happy accident, isn't it? Presumably he didn't need any AI help to do it, either. There's hope for us all then.

"The economic growth in front of us looks astonishing, and we can now imagine a world where we cure all diseases, have much more time to enjoy with our families, and can fully realize our creative potential," says Altman.

"In a decade, perhaps everyone on earth will be capable of accomplishing more than the most impactful person can today."

Well, that sounds lovely, doesn't it? Who doesn't love a promise of a Star Trek-style, utopian, post-disease, post-scarcity future, in which our endeavours are supported by our own private genius. To justify that thinking, Altman's three observations are thus:

1. The intelligence of an AI model roughly equals the log of the resources used to train and run it. "These resources are chiefly training compute, data, and inference compute. It appears that you can spend arbitrary amounts of money and get continuous and predictable gains; the scaling laws that predict this are accurate over many orders of magnitude."

2. The cost to use a given level of AI falls about 10x every 12 months, and lower prices lead to much more use. "You can see this in the token cost from GPT-4 in early 2023 to GPT-4o in mid-2024, where the price per token dropped about 150x in that time period. Moore’s law changed the world at 2x every 18 months; this is unbelievably stronger."

3. The socioeconomic value of linearly increasing intelligence is super-exponential in nature. "A consequence of this is that we see no reason for exponentially increasing investment to stop in the near future."

That last point seems particularly pertinent, given that OpenAI has previously been reported to be burning through billions of dollars in staffing and model training costs. Exponentially increasing investment has been a keystone of the modern AI boom, and the release of China-based startup DeepSeek's R1 model (supposedly trained at a fraction of the cost of existing efforts) has recently shaken investor confidence in the US-dominated AI industry.

So it's no surprise Altman is highlighting its importance here. On the first point, however, it looks like Altman has no illusions of the mainstream AI market (if such a thing exists) letting up on training costs, hardware requirements, and "arbitrary amounts of money" in order to continue gaining ground in AI development, at least when it comes to AGI.

Still, according to Altman it at least becomes cheaper to use over time. Which, if his predictions about the future of AI agents come true, will be necessary to enable our AI-co-worker hellsc... I mean, future working methods.

"Let’s imagine the case of a software engineering agent... imagine it as a real-but-relatively-junior virtual coworker. Now imagine 1,000 of them. Or 1 million of them. Now imagine such agents in every field of knowledge work.

"The world will not change all at once; it never does. Life will go on mostly the same in the short run, and people in 2025 will mostly spend their time in the same way they did in 2024. We will still fall in love, create families, get in fights online, hike in nature, etc.

"But the future will be coming at us in a way that is impossible to ignore, and the long-term changes to our society and economy will be huge. "

Goody. I'm pleased to hear that, in Altman's eyes, I'll still be getting in fights online and hiking in nature this year. But AGI-enabled assistants are coming, says the OpenAI head honcho, and given the previous trends he's highlighting here, they appear to be coming rather quickly (providing the money tap keeps flowing, of course).

"Agency, willfulness, and determination will likely be extremely valuable," Altman continues. "Correctly deciding what to do and figuring out how to navigate an ever-changing world will have huge value; resilience and adaptability will be helpful skills to cultivate."

"AGI will be the biggest lever ever on human willfulness, and enable individual people to have more impact than ever before, not less."

So, a lot of optimistic thinking going on here, it seems. I'd like to hold my hand up and say that I'm not too keen on the idea of an AI co-worker writing my articles for me, but if they could "marshall the intellectual capacity" to whittle down my inbox reliably without sending important messages to the spam folder, that'd be grand.

I don't want an AGI Mozart, more of a competent Jeeves. Still, as Altman has it, things do sound suspiciously bright and rosy for our creative futures:

"There is a great deal of talent right now without the resources to fully express itself, and if we change that, the resulting creative output of the world will lead to tremendous benefits for us all."

Here's hoping, anyway.

On an average day about a dozen new games are released on Steam. And while we think that's a good thing, it can be understandably hard to keep up with. Potentially exciting gems are sure to be lost in the deluge of new things to play unless you sort through every single game that is released on Steam. So that’s exactly what we’ve done. If nothing catches your fancy this week, we've gathered the best PC games you can play right now and a running list of the 2025 games that are launching this year.

Studio System: Guardian Angel

Steam ‌page‌ ‌
Release:‌ February 5
Developer:‌ alreti circle

Studio System: Guardian Angel is a surreal horror game that seems to borrow some stylistic quirks from Atlus. It's set in a school, for one, though more specifically, it's set in a secret film studio that just happens to be directly beneath the protagonist's school. Just as soon as this weird truth is brought to light, all students disappear and the protagonist finds themselves in this bizarre subterranea where disturbing creatures lurk. Interestingly, Studio System uses its fixed camera perspective as a tool for both combat and problem solving. Come for the dread-inducing atmosphere, stay for some interesting mechanical twists on the genre.

Slender Threads

Steam‌ ‌page‌
Release:‌ February 8
Developer:‌ Blyts

Slender Threads follows travelling salesman Harvey Green as he becomes embroiled in a small town's paranormal dramas. If I happened upon a creepy, dilapidated small town full of unutterably weird people, I would probably just leave, except Green can't: something is holding him there. Don't be fooled by the kinda-whimsical character design, because Slender Threads is a seriously unsettling psychological horror point 'n' click adventure, with some extremely sus-looking locations to explore.

While Waiting

Steam‌ ‌page‌ ‌
Release:‌ February 6
Developers:‌ Optillusion

While Waiting is a game about waiting that rewards doing nothing at all. It's a series of set pieces tracing typical "waiting" situations like waiting for the toilet, waiting for a meeting to end, or waiting in a joyride queue. It's possible to interact with all environments in granular ways, but nothing you do will rush along the thing you're waiting for. A pointless game, then? No: it's got a very obvious comedic tone, and the whole thing is wrapped up in "a thought-provoking narrative". It's probably the only game on Steam with "Easter Eggs Hidden in Boringness" listed proudly as a feature.

World End Diner

Steam‌ ‌page‌ ‌
Release:‌ February 4
Developer:‌ Andymente

World End Diner is a post-apocalyptic management game about running a diner at the end of the world. Don't be fooled by the premise though, because this is as cosy as they come, with lush 16-bit style graphics that inevitably put me in mind of Stardew Valley. Once you've gathered ingredients, whether by farming or just foraging in the woods, it's time to cook meals for the various cute animals and robots that make up your customer base. New recipes unlock as the game progresses, and there seems to be an abundance of progression paths towards making your diner the best in the world, assuming there are any other diners in the world.

Looney Landers

Steam‌ ‌page‌ ‌
Release:‌ February 6
Developer:‌ pixel games SARL-S

I was drawn to the graphics of Looney Landers: it kinda looks like something Amstrad Action would have raved about in the early 1980s. It's a local multiplayer party game about falling through a series of zany colourful obstacles without dying. The objective is to land safely, but to be competitive you'll need to land safely while taking some stupid risks. It's part of the studio's Class of 1983 arcade series, which also includes Donut Dodo and Cash Cow DX.

"Artificial intelligence is an oxymoron, there's no such thing," Zelnick said. "Machine learning, machines don't learn. Those are convenient ways to explain to human beings what looks like magic. The bottom line is that these are digital tools and we've used digital tools forever. I have no doubt that what is considered AI today will help make our business more efficient and help us do better work, but it won't reduce employment.

"To the contrary, the history of digital technology is that technology increases employment, increases productivity, increases GDP and I think that's what's going to happen with AI. I think the videogame business will probably be on the leading, if not bleeding, edge of using AI."

It's arguable that generative AI does in fact reduce employment. The maker of the Champions of Otherworldly Magic digital trading card game, for instance, said in 2024 that it paid $90,000 to a single "AI artist" for 10 hours of work because "in that time, he still makes HUNDREDS of AMAZING bits of artwork—ASTRONOMICALLY FASTER than ANY team of traditional artists." But you could also argue, as Zelnick does, that generative AI is just another tool, and part of the ongoing evolution of game development, not all that different from, say, GameMaker Studio 2.

I think that's a bit of an optimistic outlook—the whole point of automation is to remove the expense of human labor from the bottom line, after all. But if nothing else, Zelnick has been consistent on this point. He said virtually the same thing—"there is no such thing as artificial intelligence"—during a Take-Two investors call in February 2023, for instance, and in May of that year he said "artificial intelligence is an oxymoron," and that while such tools enable more efficient work, "Genius is the domain of human beings and I believe will stay that way."

Perhaps tiring of the topic, he was somewhat blunter while speaking at a TD Cowen conference in 2024. "I'm in a Whatsapp chat with a bunch of Silicon Valley CEOs, and the conventional wisdom out there is like, 'AI is gonna make us all unemployed'," he said at the time. "It is just the stupidest thing I've ever heard. The history of productivity tools is that it increases employment. It increases value, it increases yield, it enhances growth. All of these things will happen."

One interesting point Zelnick made in the GamesIndustry interview is the risk of copyright infringement that using LLMs poses. That's kind of a big thing in the world of AI development: OpenAI, for instance, recently complained that competitor DeepSeek has been using its data, which is incredibly ironic because OpenAI trains on data scarfed up from other sources, often without permission—in January 2024, in fact, OpenAI said it would be "impossible to train today's leading AI models without using copyrighted materials."

That's a mess Zelnick seems eager to avoid. "In terms of [AI] guardrails, if you mean not infringing on other people's intellectual property by poaching their LLMs, yeah, we're not going to do that," he said. "Moreover, if we did, we couldn't protect that, we wouldn't be able to protect our own IP. So of course, we're mindful of what technology we use to make sure that it respects others' intellectual property and allows us to protect our own."

Given how offputting and utterly detrimental to the experience its ads are, you might fairly wonder why YouTube keeps shovelling them on. Why does it work so hard to make YouTube worse? The answer is very simple: Those hated YouTube ads make, as they say in the Canadian scientific community, a metric ass-ton of cash.

Alphabet's Q4 financial results (via Variety) reveals that YouTube ads raked in more than $10.4 billion in the quarter—to be clear, that's 10 billion smackers in just three months. That's a relatively small slice of the total Alphabet pie, which topped $96 billion (again, in three months), but it's a staggering amount of money for three months of ads on the 'Tubes. It's also a new record, marking a 14% increase over YouTube's advertising revenue in Q4 2023, which hit $9.2 billion.

In its earnings call, Alphabet CEO Sundar Pichai attributed much of that jump to the US election, which took place on November 5, 2024. Combined spending on YouTube ads by the Democratic and Republican parties was almost double what they spent in the 2020 elections, and more than 45 million people watched "election-related content on YouTube" on election day alone.

It's possible to dodge the deluge of ads, of course, by signing up for the YouTube Premium subscription service. That goes for $14 per month or $140 per year, which strikes me as an exorbitant amount of money for, y'know, YouTube. It'll be a cold day in hell before I throw that kind of money at, again, YouTube, but at least a few of my compatriots here at PC Gamer—like associate editor Ted Litchfield—do pony up for it, and I'm ashamed to say they all claim it's worth it.

They're not alone in that assessment, because it's raking in the green too. YouTube subscription revenues aren't reported separately but are instead rolled into "Google subscriptions, platforms, and devices," but that's up year-over-year as well, from $10.8 billion in Q4 2023 to $11.6 billion over the same period in 2024. Alphabet chief financial officer Anat Ashkenazi said during the earnings call that "we continue to have significant growth in our subscription products, primarily due to increase in the number of paid subscribers across YouTube TV, YouTube Music Premium and Google One." YouTube TV, for the record, is a separate streaming service, unrelated to YouTube Premium, that starts at $83 per month.

Oh, but don't worry, because it'll probably get worse. "We believe that AI will revolutionize every part of the marketing value chain and, over the past quarter, we’ve seen how our customers are increasingly focusing on optimizing their use of AI," chief business officer Philipp Schindler said. "As an example, Petco used Demand Gen campaigns across targeting, creative generation, and bidding to find new pet parent audiences across YouTube. They achieved a 275% higher return on ad spend and a 74% higher click through rate than their social benchmarks."

Schindler said later that, based on Nielsen analysis, "Google AI-powered video campaigns on YouTube deliver 17% higher return on advertising spend than manual campaigns."

So there you have it: Everyone hates YouTube ads but they make a mint, so we're stuck with them and can probably expect them to burrow even deeper into our psyches in the future unless you're willing to spend 14 bucks every month to make them go away. Maybe that's not such a terrible price to pay after all.

As acknowledged in the recent Windows release health update, this security issue was opened on December 24, 2024, and 'resolved' on February 6. This issue notes that it happens with those using install media, like a CD or USB drive, to install Windows 11 files that were distributed in October and November 2024 updates.

As the Windows 24H2 update launched in October officially, if you happened to have upgraded as it launched, you will want to get the most recent build and install it. However, if you installed 24H2 via the Windows update tab, your Windows 11 will be able to get security updates as usual.

This was 'resolved', as the fix is effectively 'Don't install the Windows 11 updates that have security update issues.'

Windows security updates are used to fix vulnerabilities in Windows that can be exploited by third parties or can result in stability issues. Getting security updates is a necessary part of the software upkeep of your Windows 11 PC as going without security updates can put your rig at risk. This is why Microsoft ending support for Windows 10 at the end of the year is a big deal to those who haven't upgraded yet. The longer you go without regular software maintenance, the more risk your rig is undertaking.

Coincidentally, a brand new workaround to install Windows 11 on devices that don't fit the hardware requirements was created around the launch of 24H2. This means that if you used Rufus or Flyby11 recently to install Windows 11 on a rig incompatible with Microsoft's OS software, you will need to do a clean reinstall in order to fix it.

If you have the version of Windows 11 without future security updates, it's pretty simple to update and you can use the same method you did in the latter half of last year. Just grab the latest version of Windows 11 and install it.

Unfortunately, this does mean having to sit through the Windows 11 update screen, but you don't need to remove personal files to do a clean install so it shouldn't be a huge ordeal. Just make sure to do it soon, as the allure of putting off Windows updates is a tempting one. I'll get to it later.

This could be more than a minor inconvenience given the Microsoft has also announced the end of Windows 10 support is inked in for later this year.

The hardware in question is the TPM or Trusted Platform Module version 2.0. TPM is a security technology that protects critical data such as encryption keys and ensures the authenticity of hardware and operating systems. While it was introduced two years before Windows 11 in 2019, plenty of PC's lacked support when Microsoft debuted its latest operating system.

For that reason, Microsoft published a workaround that allowed Windows 11 to be installed on PCs with the older TPM 1.2 module. It was a fairly straight forward workaround, provided your PC had that TPM 1.2 module, and amounted to little more than the creation and enabling of a registry key.

Well, as spotted by Neowin, Microsoft has removed all mention of said workaround on the support webpage in question. It's not clear if the workaround still works or has somehow been disabled, but there's a Waybackmachine link anyways. You're welcome.

For owners of older PCs, this still isn't good news. The immediate response would be, meh, okay, I'll just run Windows 10. Who needs that highfalutin' Windows 11 upgrade?

But there's a catch. Last year, Microsoft announced that Windows 10 support would be ending later this year, on 14 October to be precise. Now, that doesn't mean Windows 10 suddenly stops working on that fateful day. But as Ian pithily said, "from then on, it won’t receive any more updates and will remain vulnerable to whichever zero-day exploit or AI-crypto-quantum hack comes along next."

It is possible that a CPU or perhaps more likely a CPU and motherboard upgrade would get you over the line when it comes to TMP 2.0 support. But it's a pity that Microsoft isn't more supportive of older PCs. After all, a decent spec PC from 10 years ago is well capable of running Windows 11 in terms of pure performance.

Alternatively, you could just use a boot tool, such as Rufus, which will allow you to download the latest Windows 11 ISO from the Microsoft website and check a box to get around the TPM 2.0 restriction. Job done.

There are other workarounds, too, outside of the workaround we're linking to up there for posterity. You can also use FlyBy11, which does a similar job. So, why Microsoft is suddenly against such actions, again, is anyone's guess.

On its own website, Microsoft says, "climate change requires swift, collective action and technological innovation. We’re committed to meeting our own goals while enabling others to do the same."

So, it would be welcome if Microsoft didn't force owners of old PCs to choose between landfill and limited security, wouldn't it?

"If it has been more than 12 months (one year) since the developers of a game have updated the game and posted an 'update' type event to tell players about it, Steam will add a note to the store page, just above the Developer Questionnaire letting players know that the game has not updated in a while. This notice will appear when either of the following criteria has been met:

• It has been more than 12 months since a build has been assigned to the 'default' branch in Steamworks.
• It has been more than 12 months since the developer has posted an 'update' type event, communicating to players about the update (This includes all three 'update' type events: Major Update, Regular Update, or Patch Notes)"

Original story: It's got its ups and downs, but I rather like Steam early access. Being able to get in on the ground floor of something, watch it develop, and support developers whose ambitions might exceed their initial budgets is a neat thing. The downside, of course, is that sometimes the games just don't get finished. They get left permanently in an alpha state, their devs vanishing in the night, never to be heard from again.

Which is an issue, certainly. Also an issue: there's not been much on Steam to stop unsuspecting new players from picking up those abandoned games. Unless it's garnered enough negative reviews to catch a punter's attention, up to now it's been very possible for someone to pick up a dead early access game on the assumption it's still in active development.

I say 'up to now' because Valve has finally tried to rectify the problem. As spotted by SteamDB, early access games that have gone a long time without an update will now have a highlighted note added to the top of the 'Early Access Game' section on their store page, pointing out how long ago the last change was and warning that "The information and timeline described by the developers here may no longer be up to date". It seems to kick in after 12 months or so; I can only find the note on games that have gone update-less for 13 or more months, and it's notably absent from, for instance, Kerbal Space Program 2.

You can see it in action on the page for long-fallow early access game Cavern Kings, where it clearly states that it last got an update all of eight years ago.

Which is all well and good, but I think the system could do with some tweaking. The game Heartbound, for instance, last got a patch to its beta branch four days ago, so it's still undergoing active development, but anyone just cursorily glancing at its info box might be put off by the note that its last update came "over 13 months ago." Steam might have to widen the net about what counts as an 'update,' since I imagine that's contrary to the way many people will interpret the warning in its current state.

Correction: The most recent update for Heartbound was to its beta branch, which it seems Steam doesn't take into consideration. This piece has been updated to reflect that.

Opera Air looks rather similar to Opera's other browsers but the main differences lie in two features: Boost, which are binaural beats-enhanced songs right in the browser itself that you leave playing while you do work (or doom scroll); and Breaks, which are meditation exercises to encourage mindfulness.

Before Opera Air, I was stressed, anxious, tired. After Opera Air, I am still all those things but I have a pretty new browser and feel just a little better. Having used Opera GX on and off for the last few months, I like many of the features of the browser but it can feel a bit cluttered thanks to the level of customisation. Opera Air, which you can download and test for yourself right now is the antithesis of this.

Opera Air is simple and clean. Once downloaded, you pick a theme and start browsing. Those themes are all very natural, like stones on grass or floating bubbles (okay, those bubbles are man-made but you get the point). In Speed Dial, which is essentially your home screen, you only start with a few pinned websites and they all fit a similar theme. Mindful, Headspace, Calm, Behance, Penzu, and Medium. All of these apps are focused on mindfulness and introspection, be it via writing or meditation.

As is Opera's MO, you can customise your home screen apps but the selection at the start isn't too shabby. There is a strange irony in the announcement for Opera Air saying it has 'no subscriptions', then adding multiple subscription paywalled apps to the Speed Dial but they're still a decent selection of websites.

The Boost feature has a solid selection of song loops, and you can customize how loud the track, ambient sounds, and binaural beats are, which means you can fine-tune your experience. The Creativity Boost sound in Boost has rhythmic water drops, which I only started noticing as I started to write this sentence. I don't quite know if they make me any more productive but it's certainly a relaxing experience. They also pause automatically when another sound comes out of your browser, making Boost quite intuitive throughout the flow of the day.

In the Breaks feature, you can customise sets of exercises or meditation. They are narrated with a voice, pause when you click away, and some of them even use your camera to register if you are correctly doing them. It's a frankly bizarre experience, but one that made me feel better afterwards. Engaging in mindfulness exercises can make you feel a bit silly, especially in an office full of people writing, but Opera Air encourages light mindfulness a surprising amount given it's, well, a browser.

Opera Air has surprised me just enough to make me want to use it for a little longer. I don't know if I can reasonably ask for any more.

Faceless AI-generated YouTube channels have become a problem as of late, so one tech YouTuber took it upon themselves to poison the data they are trained on.

In her latest video (via Ars Technica), f4mi has shown off a way to mess with the data that AI is trained on, specifically the subtitles of YouTube videos. Effectively, these bots can scrape all the information in the video by accessing the data in subtitle files, which are then synthesised with all the other YouTube videos it has taken data from.

Someone can then instruct that bot to create a new video in the style of other videos. The tools aren't quite advanced enough to animate a lifelike face and the creators won't put their own face on it. That is where the "faceless YouTube channel plague" (as popular YouTuber Kurtis Conner describes it) comes from.

"All this AI slop that's appearing on every social media platform is not made by robots trying to steal our jobs," says f4mi, "it's made by humans trying to make money using AI to launder other people's work."

A little while back a life hack popped up online. Job applicants were putting positive keywords in white between margins in document folders to trick AI bots into thinking they were more employable. The poisoning method that f4mi invented is sort of like this.

It grabs the standard subtitle text for a video, adds gibberish in between the lines, makes it invisible to the average user, and then feeds it back to YouTube. She did this by putting works in the public domain into her files and using a tool to make synonyms for many of the words so the AI doesn't spot what she's doing. A Python script simply went through and made these changes.

Unfortunately, tools that can transcribe audio get around this but a few large LLMs like ChatGPT can't. Instead, the AI would 'hallucinate' inventing ideas about the video based on what it can grasp from the enormous amount of text in its subtitles.

And this is where the '.ass' subtitle format comes from. It can be used for standard subtitles but also for animation and graphics, which allows the level of customisation needed to sneak in all of that pesky extra data. Entire books can be hidden with the file type's position tool to make both the size and transparency of unneeded text zero.

.ass is not among the supported files for YouTube videos—but you can convert an .ass file to an SRV3 file, which will then keep all of the data from your original file.

Unfortunately, f4mi's video had problems with phones crashing "due to the subtitles being too heavy". YouTube has regional subtitle files you can activate, which allows users to account for cultural differences in phrases and tone, so the poisoned track only exists in the UK subtitle track for now, which is the original .ass file. f4mi has uploaded standard files for Australia, which means AI summary tools are getting around the poisoned track and to an untouched one.

This means that plugging the video into an AI summary tool now works. However, as a proof of concept, this is a very smart workaround that means users wanting to steal your videos may have to do it the old-fashioned way, which might just be enough to deter many.

f4mi is one of a long line of YouTubers (like Drew Gooden and Hank and John Green) who have been fighting against AI plagiarism on the platform, and likely won't be the last. It's unlikely the next major anti-AI effort will have quite as good a file name attached, though.

New York-based cloud security provider Wiz has issued an advisory claiming its research wing identified a publicly accessible ClickHouse database, belonging to DeepSeek, left "completely open and unauthenticated" (via The Register).

The database was said to have been discovered within minutes of the Wiz research team's investigation into DeepSeek's cybersecurity resilience and it contained "a significant volume of chat history, backend data, and sensitive information."

Worse still, the database was so completely unprotected that it was possible to gain full database control and privilege escalation from inside the environment, with no authentication or defence mechanism present.

A potential attacker could have easily obtained plaintext passwords, local files, and proprietary data with a simple SQL command. Wiz duly informed DeepSeek of the open database, which it says was promptly secured.

As word of DeepSeek's efforts has spread throughout the tech industry, so have potential data security concerns from multiple sources. Data regulators from the UK, Italy, Ireland and Australia have all begun enquiries into the practices of the company, while OpenAI has complained that DeepSeek has been copying its models.

The US Navy has issued a warning to its members to avoid using DeepSeek "in any capacity", while the US National Security Council says it's looking into the security implications of the DeepSeek app.

AI security provider HiddenLayer claims that DeepSeek-R1 is "vulnerable to jailbreak techniques, prompt injections, glitch tokens, and exploitation of its control tokens, making it less secure than other modern LLMs."

Given the disruptive nature of DeepSeek's entry into the market, it's difficult to ascertain how many of these claims are legitimate, and how many may be reactionary attempts looking to restore some of the AI status quo.

Regardless, leaving a database wide open to be manipulated by any who may come prying is not a great look. It seems like no matter what happens next, DeepSeek will be at the top of everyone's AI concerns for a while to come.

Well, now GOG wants to enlist you in the good fight, and it's got a new classic release to whet your appetite for the struggle ahead: Dino Crisis 1 and 2 are back, baby. You can pick up each game—or both as a bundle—in their newly GOGified form right now.

They've joined the store's Game Preservation Program, meaning GOG is committing to keeping them running on modern systems basically forever, and like the Resident Evil games before them the Dino Crises come with tweaks to work on Windows 10/11, support 4K resolutions and modern controllers, and generally run without being a major pain. I've had a poke at the GOG version of DC1 myself and, friends, if this ain't the best way to play the game then it's a close second to a PlayStation and a CRT.

Which is lovely. So lovely, in fact, that I wish there were some way plebeians like me could shout at publishing execs about the old games I want to get the same treatment. Well, what do you know? GOG's also announcing an overhaul of its Community Wishlist feature. It's called the Dreamlist, and like before, it'll let users vote for games they want to see get revived on GOG.

What's different is that the Dreamlist sounds a lot more feature-rich, sleek, and forward-facing than the old Community Wishlist, which you might not even have known about if you were just a casual GOG store peruser. Alongside voting for the games you want to see back, you'll be invited to share stories and memories you have of particular games, because we are all wistful grandads now.

GOG says it can really make a difference, and that being able to point at a game with hundreds of thousands of votes (and potential buyers) can make executive ears perk up when nothing else works. "When talks with IP owners hit a wall, the Wishlist kept the conversation going," the storefront says, and hopes the dolled-up new version will encourage even more votes and give GOG even more leverage when it comes to negotiating rereleases of the classics.

On an average day about a dozen new games are released on Steam. And while we think that's a good thing, it can be understandably hard to keep up with. Potentially exciting gems are sure to be lost in the deluge of new things to play unless you sort through every single game that is released on Steam. So that’s exactly what we’ve done. If nothing catches your fancy this week, we've gathered the best PC games you can play right now and a running list of the 2025 games that are launching this year.

Mark of the Deep

Steam ‌page‌ ‌
Release:‌ January 25
Developer:‌ Mad Mimic

Mark of the Deep is an isometric metroidvania about pirates—with Souls-like trappings, of course. Shipwrecked on a weird island, protagonist Rookie needs to navigate the surreal landscape to find his scattered crew, which will involve a lot of tight platforming challenges, some light puzzling, and plenty of violent encounters that look as precision-oriented as Hades. All the usual stuff is here: there are skills to learn, weapons to find and interlocking areas. If you liked Death's Door or Tunic but wished they were more... pirate-y (or indeed, just have a hunger for any pirate game) this looks well worth a play. There's a demo.

Cat Detective Albert Wilde

Steam‌ ‌page‌
Release:‌ January 25
Developer:‌ beyondthosehills

As the name implies, Cat Detective Albert Wilde is about a cat detective—by the name of Albert Wilde—tasked with solving a "quantum mystery" that will see him utilise his "years of experience of interrogating and drinking at bars". Draped in the grainy black and white of 1930s noir, Cat Detective is a comedic, narrative driven game that dares wonder aloud what might happen if cats could be detectives and grapple with quantum theory. I for one am keen to see the results.

Summer Islands

Steam‌ ‌page‌ ‌
Release:‌ January 26
Developers:‌ Marttalin Games

Summer Islands is about running a picturesque island resort. First you shape your island—in a way that is pleasantly reminiscent of Sim City 2000—and then you start building your resort on it, which can be as big or small as you like. This being an island resort, you'll want it to be well-serviced with all the usual things tourists like, such as discos and golf courses, and once you've put everything into place you'll need to make sure everything functions and that tourists are happy. The goal is to get extremely rich, after all. Sounds like a lot of hard work, but if it's a new management obsession you're after, I reckon this is worth a shot: I love the art style in particular.

Streamer Life Simulator 2

Steam‌ ‌page‌ ‌
Release:‌ January 24
Developer:‌ Cheesecake Dev

Cheesecake Dev pumps out simulators at an alarming rate (see also: Internet Cafe Simulator, Dating Simulator and Hookah Cafe Simulator) but they tend to find their audience. This sequel to Streamer Life Simulator is a case in point: despite being launched into a hilariously crowded simulator market, it already has over 600 "very positive" reviews. The trailer above has it all: dangerous driving in the wilderness, for some reason, house cleaning, cat robbery, and photography (featuring an inflatable pool, of course). It's an early access affair but apparently there's plenty of stuff to do already. If you've played these games a bunch, can you email me?

Space Engineers 2

Steam‌ ‌page‌ ‌
Release:‌ January 28
Developer:‌ Keen Software House

Launched into early access last week, this sequel to the popular space exploration and survival sim seeks to do what all good game sequels do: be a bigger and better version of the original. As such, this alpha build is a tad barebones at the moment, because it's basically just a creative mode. So unless you've already sunk countless hours into the original, maybe try the predecessor first. For anyone who sank their teeth deep into the first Space Engineers, this definitely looks like an impressive step up.

DeepSeek was founded in 2023 by Liang Wenfeng, the co-founder of High Flyer, an AI-driven hedge-fund (via Bloomberg). Despite being a mere fledgling in the AI world compared to some of the bigger names like OpenAI and Meta, the DeepSeek V3 family of AI models appear to be much more efficient than competitors' efforts like ChatGPT.

Now R1, a new reasoning model fine-tuned from V3 training, also appears to be comparable in performance and accuracy to OpenAI's efforts.

The Register has performed some early testing on R1 in comparison to OpenAI's o1, a similar "Chain of Thought" (CoT) model, and found that its capabilities seem remarkably similar—with the former even beating out the OpenAI model in the MATH-500 test.

All while costing much less to train, with a Chiphell forum user claiming that "DeepSeek-R1 is priced at only about 3.65% of OpenAI o1" per million outputs.

That's potentially bad news for Nvidia, which has been raking in cash as a result of high sales of its AI accelerator hardware to run the latest models at scale, taking the company's valuation up to the highest by market cap in the world at time of writing.

According to some analysts, Nvidia's stock price is on track to shed nearly $400 billion dollars in the wake of R1's launch, although it appears to be holding relatively steady after a slump in early Monday trading. Dutch chip-maker ASML has seen its share price fall by more than 10% since R1's launch, while Siemens Energy dropped by 21%.

US venture capitalist and advisor to Donald Trump, Marc Andreessen, made the comparison of R1's launch to the US and USSR space race, calling it "AI's Sputnik moment." Sputnik was a satellite launched by the USSR in 1957, and is widely thought to have wrongfooted the US government on the Soviet Union's space capabilities, leading to a space race between the two nations.

The elephant in the room, however, is trade sanctions. The US government has banned the export of high-end AI hardware to China, which some have speculated has led to a more efficient approach resulting in DeepSeek's new models. If so, that would suggest that the sanctions haven't had the effect that the US gov may have hoped.

However, being a Chinese product (and given that many of the leading AI companies are US-based), it seems possible that restrictions may follow on its use for US companies, and potentially even those using US equipment, in the same fashion as the current trade sanctions.

Either way, the release of DeepSeek-R1 appears to have caused a significant rumble in the AI markets. Whatever happens next, it seems that the US may no longer hold all the keys to advanced AI—and the world awaits it's response.

And further down the post, under known issues? "The Home page of settings may crash." I've yet to receive the update myself, but here's how my homepage looks currently, with multiple cards already present.

While it's not clear if the new ad is to blame for the crashing issue, I can't be the only one increasingly annoyed by the fact my Windows settings are becoming cluttered—especially as I am a Game Pass subscriber, and I don't need new features advertised to me for something I'm already paying for when I'm scrolling my settings menu.

Alright, I'm not really on the homepage that often, and you probably aren't either. But for the less PC-savvy, greeting settings explorers with a bunch of ads for other Microsoft products and features doesn't exactly make the experience straightforward—especially if the homepage crashes before they find the setting they're looking for.

Bloat is bad, and bloat this most certainly is. Anyway, there are still some useful features added in the new update, including the ability to use your Xbox controller to navigate and type on a new Gamepad keyboard layout, some new Narrator functionality, and some explainer messaging added to Snap.

Microsoft says this is aimed at "driving contextual learning and discovery", which essentially amounts to some text at the top of Snap hover windows telling users what each function does.

It's not the most exciting of updates, granted. But at least here Microsoft seems to be making an attempt to make Windows features easier to navigate, rather than shoving yet another ad in to important pages that are fairly cluttered to begin with, potentially breaking them in the process.

Personally, I'm still waiting for a build that fixes the litany of issues created by the Windows 11 24H2 update. In the meantime though, some accessibility is added—and some, it seems, is buried under yet more ad cards. MS giveth, and MS advertises its products in your settings menu. The world keeps turning, folks.

An anonymous coder, writing to 404media under the fake name Aaron B, has also been feeling the malaise, and he's doing something about it. Aaron B has created a digital flytrap to keep web-crawling bots in a thus-far undetectable doom loop, scanning the same pages over and over until someone human comes over and fixes it.

The program, dubbed Nepenthes (after the tropical pitcher plant), exploits a weakness in these crawlers—the fact that they're apparently really dumb, as Aaron B explains:

"It's less like flypaper and more an infinite maze holding a minotaur, except the crawler is the minotaur that cannot get out. The typical web crawler doesn't appear to have a lot of logic. It downloads a URL, and if it sees links to other URLs, it downloads those too. Nepenthes generates random links that always point back to itself—the crawler downloads those new links. Nepenthes happily just returns more and more lists of links pointing back to itself."

The site itself lays it out in full, as well as the usage cases—allowing you act defensively to "flood out valid URLs within your site's domain name, making it unlikely the crawler will access real content." You can also, as the site suggests, play offense by ignoring the code's existing list of known trawler IPs and allowing bots to "suck down as much bullsh*t as they have diskspace for, and choke on it."

There's also a link to a demo of the software in action, as well as an all-caps warning that "THIS IS DELIBERATELY MALICIOUS SOFTWARE INTENDED TO CAUSE HARMFUL ACTIVITY" and that you will experience "significant CPU load" if you use it. Govern yourselves accordingly.

This isn't a perfect solution to the plague of AI scrapers—most AI models guzzle a huge amount of fuel, and Nepenthes doesn't do anything to stop that: "They are still consuming resources, spinning around doing nothing helpful, unless they find a way to detect that they are stuck in this loop."

Essentially, there's sort of a dark feedback loop that's possible here—these scraping bot farms we're all very tired of could be motivated to add more scrapers to counteract them, driving up energy costs and… you know, accelerating the horrors. It's very much a 'fine, we'll both burn' stopgap. Aaron B even couches it in such terms:

"It's also sort of an art work, just me unleashing sheer unadulterated rage at how things are going. I was just sick and tired of how the internet is evolving into a money extraction panopticon, how the world as a whole is slipping into fascism and oligarchs are calling all the shots—and it's gotten bad enough we can't boycott or vote our way out, we have to start causing real pain to those above for any change to occur."

As for the claims that bots can skip over these traps, Aaron B seems confident he's at least wasting some of their time: "I've several million lines of access log that says even Google Almighty didn’t graduate."

Examples of the sorts of things Operator can do are booking travel, making restaurant reservations for a certain time, or perhaps buying concert tickets for a specific band within a given price range.

Currently released as a research preview only available to ChatGPT Pro subscribers rather than a fully baked product, Operator is based on OpenAI's Computer-Using Agent (CUA) model, which combines the computer vision capabilities from GPT-4o's with specific graphical user interfaces (GUIs) training and advanced reasoning to create a tool capable of browsing the web, formulating multi-step tasks from a text prompt and executing the whole shebang.

Arguably, Operator is not unique, what with ByteDance's UI-TARS and Anthropic’s Computer Use having a somewhat similar remit. But perhaps what makes Operator a little different is that it doesn't need APIs.

"Operator can 'see' (through screenshots) and 'interact' (using all the actions a mouse and keyboard allow) with a browser, enabling it to take action on the web without requiring custom API integrations,' OpenAI says.

That said, it does seem like it helps if web services are optimized for Operator. "We’re collaborating with companies like DoorDash, Instacart, OpenTable, Priceline, StubHub, Thumbtack, Uber, and others to ensure Operator addresses real-world needs while respecting established norms," OpenAI says.

Presumably, your results—or should that be Operator's results?—won't be as accurate with non-optimized services.

Exactly how good Operator currently is at taking a prompt and running with it isn't entirely clear. OpenAI itself says the CUA models returns a 38.1% success rate on the OSWorld benchmark test for full computer use tasks, a 58.1% for WebArena, and 87% on WebVoyager for web-based tasks.

Meanwhile, some early users are reporting that Operator may be more prone to hallucinations than recent builds of ChatGPT itself. For instance, one user claims that when tasked with generating a list of online influencers and tabulating their contact details, Operator entirely made all the details up.

Some users also report that Operator is surprisingly slow, something that seems to tally with the video demo posted by OpenAI. It's hardly a whirlwind of mouse inputs, that's for sure.

So, it seems you'd be brave to ask Operator to do your grocery shopping next week and expect all the right stuff to turn up, or perhaps anything to turn up at all.

The big question, then, is how quickly this early beta version of Operator will develop into something broadly reliable and useful. And then of course how safe it will be if and when it does.

"We know bad actors may try to misuse this technology. That’s why we’ve designed Operator to refuse harmful requests and block disallowed content," OpenAI says, also explaining that Operator has been designed to deal with websites that might try to hijack the AI agent with hidden prompts, malicious code, or phishing attempts.

For now, these are all, ultimately, imponderables. But for better or worse it does seem like you'll soon be able to hand over quite a few of your routine online tasks to an AI agent.

It's here, what we've all been on the edge of our seats for: Game Assist in Microsoft Edge Stable. Okay, sorry, sorry—for the joke and for the sarcasm. But yes, Microsoft's foray into anti-alt-tabbing activism is now out of beta.

For the uninitiated, Game Assist debuted in November last year, much to the seeming chagrin of our Rich Stanton, who couldn't figure out exactly who the target market is for the feature. It's essentially a browser that you can open without alt-tabbing, via your standard Win+G Game Bar.

The key allure, though, apart from it attempting to find relevant guides and information for whatever you're playing in select games, is its ability to be pinned on-screen while gaming. Y'know, for all the single-screen peasants (kidding—just a personal jab at my single-screen defenders here in the PC Gamer hardware den).

One big problem with it, though, was that it required you to change your default browser to Edge or install some Insider software. But now that it's been released in the Stable channel (i.e. the public release), you don't have to do that; you can just opt-in by installing the Game Assist widget in your Edge browser settings.

Once you do this, you'll be able to use Game Assist while gaming by hitting Win+G to open your Game Bar. If you're in a supported game—which now includes a handful of new titles such as Marvel Rivals and Dragon Age: The Veilguard—it should attempt to grab you relevant info or guides for whatever you're playing.

Otherwise, you can just use it like a normal browser and then pin a YouTube guide video (for example) to the top-right of your screen while you frantically attempt to keep up with the guider's speedy instructions.

I must say, I don't share the same bafflement as some others regarding this feature. I remember watching some streamers play games while using a third-party app to pin videos to their screen to watch at the same time. Which certainly seems like a good shout for those with somewhat of a deficit of attention. I used to enjoy watching podcasts while grinding in an MMO, for instance.

Who knows, maybe this will make me actually use Game Bar for once. That is, for something other than attempting to join a friend's party when in-game social options are confusing, only to immediately regret it and spend five minutes half-absentmindedly but frustratedly trying to figure out how to leave the abominable party or at least mute my teammate to prevent doubled voice comms.

Yeah, not a fan—can you tell? Only time will tell whether Game Assist can change my mind.

The just-released Windows 11 Preview Channel Build 26100.3025 (KB5050094) has a few new features and improvements, but given the as-yet buggy 24H2 experience, I'm more focused on those "fixed" notes, of which there are a fair few.

This build is an Insider one, meaning it's an early version that you can access by signing up for the Windows Insider Program, but these builds do hit the main release channel eventually.

You can check out the full list of fixes yourself on the Build announcement, but here are some of the bugs that we're particularly happy to see getting fixed:

• A disappearing mouse cursor when you hover over certain text fields in some applications.
• The mouse cursor stuttering when it moves across the screen.
• Oversaturated HDR in some games when you use Auto HDR.
• Searches repeating for no reason.
• The search box losing input focus as you type.
• Task Manager taking a long time to close.
• Snipping Tool screenshots being distorted.

Here in the PCG hardware den, a bunch of us have faced such kinds of issues with 24H2, from Auto HDR game crashes to audio device issues, stuttering, and more. I've personally faced animation glitches when minimising and maximising windows, Nick's had 24H2-induced graphical issues in Chrome, and Andy seems to have had the worst of it, with broken drivers, disappearing and reappearing notifications, and in-game stutters and hangs.

Some of us have also experienced mouse cursor stutters, which seems like the most irritating and ever-present bug of the lot. And also possibly the most baffling issue of them all. It's 2025, so how is it that Microsoft rolled out an update that breaks one of the most fundamental things to get right for GUI operating systems? (The 'Small Indie Company' joke is fruit that's almost too low-hanging to mention.)

Microsoft's reasoning might be that 24H2 is apparently a "full operating system (OS) swap". This sounds silly, given the user experience is mostly the same but with some AI gubbins sprinkled here and there. But I suppose we're only getting the front-end experience; a complete low-level redesign might have been needed to get Windows ready for the AI PC era.

Whatever the reason, and however ridiculous some of the bugs may seem, at least new builds are attempting to fix them. Let's just hope new bugs aren't thrown into the mix for good measure—if there's anything that'll be sure to keep the Windows 10 die-hards from updating, it'll be that tiresome whack-a-mole.

Order 14110 was issued in October 2023 and aimed to reduce the risk associated with developing AI technologies. It required AI system developers to share the results of safety tests with the US government and directed relevant federal agencies set standards for those tests.

The order, originally titled "Safe, Secure and Trustworthy Development and Use of Artificial Intelligence," also contained provisions for the protection of copyright material involved in the process of training AI models, such as ChatGPT, set out a framework to minimize "unlawful discrimination" due AI model bias and sought to prevent US jobs from being harmed by AI.

When the order was issued, it drew support from some inside the AI industry, with Microsoft President Brad Smith calling a "critical step forward in the governance of AI technology."

But whatever the pros and cons of the order, one thing we can say for sure is that it's officially a goner. Among a list of fully 78 other Biden era orders deemed "harmful" by the new administration, Order 14110 has been immediately revoked.

Indeed it's hard to know exactly what impact this will all have. On the one hand, it seems to open the floodgates for an unregulated AI free-for-all. On the other, it's unclear how much impact the now-cancelled order had. In practice, its revocation may make little difference.

That said, it's notable that the EU has its own AI Act, signed into law in March 2024, which bans the use of biometric data, requires new high-risk systems to undergo safety testing before release, and insists that copyright data used in training must be disclosed.

Moreover, leading figures in the industry such as OpenAI's Sam Altman have called for greater regulation and the creation of a new federal agency tasked with licensing operators who breach a certain threshold of AI capability.

At the same time, it's likewise clear that even the most advanced and best-resourced tech companies are capable of releasing AI tools that either don't work very well or raise safety concerns, whether its the on-again, off-again debacle with Microsoft's Recall for Windows or the more recent suspension of Apple's AI news summaries.

Ultimately, it's hard to believe that the development and use of AI systems will be safer with the guardrails taken away. So, the tradeoff here is a balance between the economic and strategic benefits of taking the AI industry in the US off whatever leash was constraining it, versus any damage that might be done by releasing unregulated AI tech.

For the record, the White House URL for the original order now 404s, so the Trump administration has fully purged it from existence. All that's really left to do is buckle up, cross your fingers and toes and see how the next four years go.

But TeamSpeak never went away, and it looks like 2025 may be the year of its triumphant return. Or at the very least, a very good attempt to give Discord a run for its money. The TeamSpeak X account has been doing some pretty cryptic posts since the beginning of the year, like "Man I love January, such a great month, not saying anything is going to happen this January, but wow, it's great. January btw."

Vagueposting has now turned into just straight-up posting, however, with TeamSpeak slowly unveiling its new look. In its own typical TeamSpeak way, of course. The new UI is looking a lot friendlier to use, putting itself just close enough to Discord's that it'll be nice and familiar if anyone does end up moving over. It looks like there are group chats, servers, and bookmarks for easy navigation.

A follow-up image shows off a little bit more, with community servers for a bunch of different games like Minecraft, Diablo 4, and Overwatch 2. TeamSpeak says the new design overhaul will include screen sharing, camera sharing, 1440p streaming, multi-streaming, all in 60 FPS. Oh, and one secret final thing apparently, though we'll have to wait a little longer to find out what it is.

I say the unveiling has been in TeamSpeak's own unique way, because anyone who's followed the X account knows that the company isn't afraid of throwing a bit of shade at Discord… pretty much all the time. It's been following up each post with a link pointing towards Discord's support site: One link on how to delete your Discord account, and another on how to cancel your Nitro subscription.

I'm all here for a little bit of brand-on-brand violence, and honestly I hope TeamSpeak manages to throw itself back into the limelight a little. Competition is always a good thing, and I've not exactly been quiet about my growing frustration with Discord and its assortment of unnecessary features and changes. But I do wonder how on board folks will be with the whole 'paying to run a server' thing that it has. Unless that's the huge final change that's coming, mind you. We'll have to wait and see.

The tool in question is called Expresso because…umm, I'm not sure why, to be honest, but it's by EasyStudio AI which also offers an AI for editing and tweaking photos. I should imagine that there was a meeting of managers, at some point, that wondered what else its servers could be used for and they all settled on creating something to make custom headshots.

Although the tool's interface isn't overly intuitive, it is simple to use—just upload a headshot photo or image, and then select a preset for a particular expression you want the face to have. You can adjust the positions of the head, eyes, eyebrows and mouth to get that perfect 'smile that never reaches the eyes' or 'ermahgerd it's so bad' look that's pretty much the norm on every YouTube thumbnail these days.

Being the inquisitive and top investigative journalist that I am, I signed up for a free account (which gives you three goes at generating an image, in the form of credits) and set about seeing just what AI could do for me. Before we get on to how insufficient three credits are at road testing something that would otherwise cost $10 a month for 100 credits, let's see what I managed to create.

Look on my works, ye mighty, and despair! It's almost as if Percy Shelley was a 19th-century AI prophet, rather than some genteel British poet. Anyway, I have to say that Expresso somewhat lives up to its name, simply because it spits out an image very quickly, though I dare say that if it becomes super popular and the servers clog up with requests, it might not be quite so speedy.

However, my issue with it is that there's no kind of preview system and you will use up one credit to generate any image. If you're a full-time content creator and you're looking to create the perfect thumbnail clip, then you could well burn through a whole stack of credits just for one headshot.

The first AI-generated image I created with Expresso (second in the gallery above) was with the standard 'surprised' preset but the final two are tweaked versions of 'shocked' and 'confused' respectively. I don't think I'm alone in thinking that none of these facial expressions (oh wait, that's why it's called Expresso…express-o-ions…) are good approximations of those emotions, but at least they look realistic enough.

But that's AI to a tee and anyone who's spent time with an AI image generator, such as Stable Diffusion, will know that you have to spend a lot of time trying small changes over and over again, to get the exact image you want. In the case of Expresso, that means a whole lot of money.

I'm clearly not the target audience for this AI tool but I'm not overly convinced that any professional YouTuber will spend good money on this when it's not really all that hard to have someone take a few shots of you with a camera and then quickly edit them in a free photo editing app.

AI doth giveth, AI doth taketh awayeth.

Star Wars Outlaws, Avatar: Frontiers of Pandora, and Assassin's Creed: Odyssey, Origins and Valhalla are said to be fixed, after Microsoft previously prevented the update from being applied on machines with those particular games installed. The Windows 11 24H2 known issue page has been updated to reflect the resolved status as of 17 January. According to MS:

"Star Wars Outlaws and Avatar: Frontiers of Pandora are no longer affected by this issue due to temporary hotfixes deployed by Ubisoft. While these fixes mitigate the crashes, players might still experience some performance issues.

"Assassin’s Creed: Odyssey, Origins, and Valhalla are no longer affected by this issue due to fixes deployed by Ubisoft. These updates have resolved the compatibility issues with Windows 11, version 24H2, ensuring that players can now enjoy these games without previous disruptions.

"The safeguard hold (ID 54437462) that was previously in place has been lifted as of 1/16/2025 to exclude these games."

Just as well really, as the Windows 11 24H2 rollout appears to have shifted up a gear. Windows Latest spotted confirmation that the troubled update is now automatically downloaded and installed when Microsoft considers your device to be compatible.

That's for Home and Pro SKUs, anyway. Enterprise users aren't on the list just yet, and the rollout is still staged, so it might be a while before it turns up on your home PC.

Still, it's coming, so we can only hope that more of the ongoing issues are resolved in a relatively timely fashion. Several of us on the PC Gamer hardware team have experienced multiple issues since the installation of 24H2, ranging from game stuttering to odd mouse behaviour and stacked notifications, so it appears to still have the odd teething issue worth keeping an eye out for.

That's a major Windows update for you, I guess. All of them seem to come with a shopping list of various issues, and giving the complexity of Microsoft's OS and its growing feature set at this point, I suppose it should come as little surprise.

This particular update seems to have caused more issues than most, though, so here's hoping Microsoft squashes further bugs wherever it finds them. Or at the very least, external developers keep on top of issues borking their products, anyway.

On an average day about a dozen new games are released on Steam. And while we think that's a good thing, it can be understandably hard to keep up with. Potentially exciting gems are sure to be lost in the deluge of new things to play unless you sort through every single game that is released on Steam. So that’s exactly what we’ve done. If nothing catches your fancy this week, we've gathered the best PC games you can play right now and a running list of the 2025 games that are launching this year.

Love, Internet, and Murder Magic

Steam ‌page‌ ‌
Release:‌ January 14
Developer:‌ Renka

Love, Internet and Murder Magic is a visual novel roguelike about managing an AI streamer. Though an AI, this streamer is based on a real person who was murdered by a serial killer, which adds a kinda disturbing dimension to the whole enterprise. As the AI streamer's agent, it's your responsibility to manage her online presence, make hard decisions about what kind of material you want to publish, and all the while try to figure out who the serial killer responsible for the real person behind the streamer is. It's a fascinating concept—I've never seen a roguelike VN before—and it looks like it wanders some psychologically discomforting territory.

Cyclopean: The Great Abyss

Steam‌ ‌page‌
Release:‌ January 18
Developer:‌ Schmidt Workshops

If Skald: Against the Black Priory looked too modern for you, perhaps Cyclopean: The Great Abyss will appeal? It's a "blobber" dungeon crawler with a charming monochromatic art style, and like Skald, it sips heavily from the Lovecraft chalice. Indeed, Cyclopean is set in the underworld of Lovecraft's Dreamlands, a setting that recurs across a handful of the author's stories. This is an Early Access affair: during its estimated six months in development it'll get more quests, more complex interactions, and an expanded sanity system.

Tales of Graces f Remastered

Steam‌ ‌page‌ ‌
Release:‌ January 17
Developers:‌ Tose Co., Ltd.

Tales of Graces f has finally been rescued from its PS3 exclusivity. I played this back in 2010 and don't remember a single thing about it, but I think that's par for the course when it comes to the Tales of... games: they're staunchly traditional JRPGs with a cosy vibe and tropes galore. Compared to some of the Tales games that have come since, Tales of Graces f is a lot more cartoony and, well, unavoidably childish, but I know there are some people who have been waiting for this to be saved, and so here it is.

Tyrant's Realm

Steam‌ ‌page‌ ‌
Release:‌ January 17
Developer:‌ Team Tyrant

Tyrant's Realm is a roguelite soulslike with a jagged 'n' blocky PS1 aesthetic. That basically means: tense third-person melee combat in ever-changing dungeons with unremittingly bleak vibes. Being a roguelite, there's some macro-progression, mostly in the form of unlockable weapons and armor (think Hades, I guess). I'm always wary of any game that relies heavily on soulslike combat: very few studios can get it right. But Tyrant's Realm is, at least, receiving some very positive reviews on Steam, so it's probably worth a spin. There's a free demo too.

Dragon Ruins 2

Steam‌ ‌page‌ ‌
Release:‌ January 17
Developer:‌ Graverobber Foundation

This sequel to last year's ultra-minimalist blobber RPG is ever-so-slightly more graphically advanced than its predecessor: the walls are a high-tech cream colour. Though I partially joke, it does look a lot more appealing than the first Dragon Ruins, which was positioned as a "dungeon crawler for tired people". The new one is described the same way, meaning this would be a fun coffee break style blobber for people who don't want to spend a fortnight reading a manual before embarking on an inscrutable stats-crunching adventure. It's casual, in other words, but dripping with atmosphere.

The malware in question is known as PlugX, and it's a particularly nasty remote access trojan that's believed to have been around since 2008 and is said to have been a favourite tool of a hacking group referred to as "Mustang Panda"(via Gizmodo). The malware receives commands via a control server, one of which the FBI gained access to with the help of the French authorities in order to identify the IP addresses of PlugX-affected computers.

Once the infected PCs were identified, the FBI then sent commands of its own via the now-compromised server, instructing it to delete itself remotely.

4,285 US machines were healed in this manner, the FBI says, with many thousands more cleansed in a similar fashion by partner law enforcement agencies around the world.

If this were a hacker movie under my direction, however, here's the point where I'd cut to a shadowy room and an ominous soundtrack.

While this is certainly a victory for the authorities, the likelihood is high that PlugX infections are much more widespread than indicated by this relatively small batch. Cybersecurity firms have long been aware of widespread usage of the trojan, and it's estimated that roughly 2.5 million devices were infected back in 2024.

These infected devices were discovered by pinging outwards from a different command-and-control server—and the data suggests that, far from being nipped in the bud, PlugX may well be chugging away in the background of a significant portion of PCs as we speak.

So, a small victory in the grand scale of things, it seems. That being said, the method of attack here has a beautiful simplicity to it. It reminds me of James Bond movies, where the evil villain's lair has a big red button marked "destroy my nefarious plan instantly" that our noble hero must endeavour to press at all costs.

Except this time, it involved keyboards and code prompts rather than Walther PPKs and a tricked-out Aston Martin. Not quite as glamorous is it? Still, vodka martinis all round, I reckon. The world is ever so slightly safer today, and in these troubled times, I'll take it.

Apparently, more than half a million TikTok users have "recently" left the platform in favour of RedNote, all in protest at the imminent US ban. “Our government is out of their minds if they think we’re going to stand for this TikTok ban,” a user called Heather Roberts reportedly said in a video message on RedNote.

Speaking of the ban, the shizzle there is that the US Supreme Court is expected to uphold an earlier lower-court ruling. If upheld, the ban would go into effect on January 19—unless, that is, Chinese owner ByteDance sells TikTok's Chinese assets, though ByteDance has said such a divestiture "is simply not possible: not commercially, not technologically, not legally."

Meanwhile, RedNote very roughly is a Chinese analogue to Instagram with added search engine aspects. Its Chinese name, Xiaohongshu, translates to "Little Red Book", which commonly refers to the famous (or, you might say, infamous) collection of utterances by Chinese Communist leader Mao Zedong.

More recently, RedNote has moved into live streaming, too. According to Reuters, it has more than 300 million monthly users, which is pretty big, even for China.

Anyway, by some accounts, the mass exodus has seen US and Chinese social media users connect like never before. CNN says that most Chinese users have warmly welcomed their new US RedNote siblings.

"It feels like so much has changed in an instant. Ordinary people from our two countries have never really connected before," CNN reports one Chinese user as commenting.

As heartwarming as all that is, it seems a little unlikely that the bonhomie will carry on indefinitely, not least because of similar sentiments shared on the Clubhouse app back in 2021 before Chinese censors stepped in.

As a non-TikTok user with absolutely no skin in this game, it's all a little baffling. What to make of this "protest" or the fact that users are willingly jumping onto another Chinese app?

It's likewise hard to predict what might happen with the incoming Trump administration. Fair to say the once-again President hasn't been entirely consistent with his attitude to TikTok. But then again, Mao himself said, "Contradiction and struggle are universal and absolute." He got that much right, that's for sure.

On an average day about a dozen new games are released on Steam. And while we think that's a good thing, it can be understandably hard to keep up with. Potentially exciting gems are sure to be lost in the deluge of new things to play unless you sort through every single game that is released on Steam. So that’s exactly what we’ve done. If nothing catches your fancy this week, we've gathered the best PC games you can play right now and a running list of the 2025 games that are launching this year.

S4U: Citypunk 2011 And Love Punch

Steam ‌page‌ ‌
Release:‌ January 9
Developer:‌ U0U Games

S4U is a text-based narrative game set in the far distant past of 2011. Like Hypnospace Outlaw, it's one of those games that takes place on a fictional PC desktop. Protagonist Miki works as a point of contact for her employer, an architecture firm, and it's partially your job to field inquiries. But you'll also be interacting with other software, engaging in non-work conversations of a much more personal nature, working on architectural blueprints, and even meeting up with people IRL. It's a gorgeous looking thing, evoking a not-too-distant past that already feels like a completely different world.

Reviver

Steam‌ ‌page‌
Release:‌ January 9
Developer:‌ Cotton Game

Reviver is a narrative puzzle game with meticulously detailed hand-drawn art. It follows the entire life of its two main characters, Carlos and Felicia. While puzzle solving is a huge element (there are over 50 puzzles to complete), the decisions you make throughout will affect the relationship between these protagonists too. Reviver comes from the same studio who brought us Sunset Hills, a similarly atmospheric narrative-puzzler.

Freedom Wars Remastered

Steam‌ ‌page‌ ‌
Release:‌ January 10
Developers:‌ Dimps Corporation

Freedom Wars originally released for the PlayStation Vita in 2014 and was hugely popular in Japan (as far as Vita games went, anyway). It's a third-person action RPG with Monster Hunter trappings, though its sci-fi setting kinda puts it closer to something like God Eater. Freedom Wars has a reputation for being a little tonally bonkers, and while I'd definitely recommend playing it, you should probably go into this with its handheld origins front of mind. Still, in addition to being a much smoother and sharper experience, this remaster also overhauls the weapon crafting system and adds some more difficulty options.

Sea Fantasy

Steam‌ ‌page‌ ‌
Release:‌ January 7
Developer:‌ Metasla

How can you pass up a 16-bit style RPG centred entirely around fishing? Sea Fantasy is exactly that: you roam a pleasant open world looking for great fishing spots in your pursuit of "SeaAZ"—which are basically fish. There's also puzzle-centric dungeon crawling and fishing rod crafting. Most enticing is the promise that this is not a "peaceful" fishing game. No: this is a tense, dramatic, nail-biting fishing game with apocalyptic undertones.

Cursed Digicam

Steam‌ ‌page‌ ‌
Release:‌ January 10
Developer:‌ Chilla's Art

Chilla's Art strikes again with another single-sitting horror experience. As the name implies, this is about a "cursed digital camera" that "captures ghosts from a strange website". If you capture a photo of a spirit, uploading it to a mysterious website will liberate them. As peaceful as that sounds, I've got a strong feeling that something really bad will happen in Cursed Digicam.

Facebook has implemented a variety of "complex systems" over the years to address the spread of harmful material online, Zuckerberg said in a video message, but those systems sometimes make mistakes—and it's now reached a point where "it's just too many mistakes and too much censorship." He also believes the recent election of Donald Trump as president of the US represents "a cultural tipping point towards once again prioritizing free speech," which he said has been under fire in recent years from governments and "legacy media."

"We're going to get rid of fact-checkers and replace them with Community Notes, similar to X, starting in the US," Zuckerberg said. "After Trump first got elected in 2016, the legacy media wrote nonstop about how misinformation was a threat to democracy. We tried in good faith to address those concerns without becoming the arbiters of truth. But the fact-checkers have just been too politically biased and have destroyed more trust than they created, especially in the US."

Facebook and Instagram will also simplify their content policies to "get rid of a bunch of restrictions on topics like immigration and gender that are just out of touch with mainstream discourse," and that have "increasingly been used to shut down opinions and shut out people with different ideas." Enforcement policies will be changed to focus on "illegal and high-severity violations," while "lower-severity violations" will require complaints before any action is taken. Zuckerberg acknowledged that the change means Facebook and Instagram will "catch less bad stuff," but said reducing the number of improper takedowns is a greater priority.

I don't think anyone would claim that Facebook's content moderation policies have been particularly great (or even effective) until now, but following in the footsteps of X is an unexpected choice. The platform originally known as Twitter has descended into near-complete chaos and uselessness since Elon Musk's takeover, which among other things saw the elimination of content moderators and its Trust and Safety team.

The net result of that farcical commitment to free speech, as PC Gamer's Joshua Wolens put it so eloquently, is that using the platform has "become dank and ichorous, like wading through the sheer, concentrated sludge of all the worst comment sections on the internet. Scam after scam after scam after AI video after crypto pump-and-dump. A pure, unmitigated firehose of slop, not suitable for human life."

Bizarrely, Meta's trust and safety and content moderation teams will be moved out of California and into Texas, where Zuckerberg said there will be "less concern about the bias of our teams." That too may be inspired by X, which moved its own headquarters from California to Texas in September 2024 and a month later changed its terms of service to direct disputes to the US District Court for the Northern District of Texas—a "favored destination for conservative activists and business groups," according to The Guardian—even though X itself is actually headquartered in the Western District.

Musk reacted to Zuckerberg's with a brief message posted to X: "This is cool." X CEO Linda Yaccarino also expressed support for the change, writing, "Fact-checking and moderation doesn't belong in the hands of a few select gatekeepers who can easily inject their bias into decisions. It's a democratic process that belongs in the hands of many. And as we've seen on X since Community Notes debuted, it's profoundly successful while keeping freedom of speech sacred."

Zuckerberg's supplication in the face of Trump's promise to punish his perceived enemies isn't subtle, but neither is it surprising: Companies and CEOs including Meta, Amazon, Apple CEO Tim Cook, and OpenAI chief Sam Altman, have donated millions of dollars to Trump's inauguration fund, a bit of ritualized self-debasement described by NPR as "kissing the ring."

Still, just in case it might go missed, Zuckerberg also worked in a little MAGA-themed commentary on Meta's planned approach to regulation of its platforms outside the US.

"We're going to work with President Trump to push back on governments around the world that are going after American companies and pushing to censor more," Zuckerberg said. "The US has the strongest constitutional protections for free expression in the world. Europe has an ever-increasing number of laws institutionalizing censorship and making it difficult to build anything innovative there. Latin American countries have secret courts that can order companies to quietly take things down. China has censored our apps from even working in the country.

"The only way that we can push back on this global trend is with the support of the US government. And that's why it's been so difficult over the past four years, when even the US government has pushed for censorship. By going after us and other American companies, it has emboldened other governments to go even further. But now we have the opportunity to restore free expression, and I am excited to take it."

Meta's content moderation cutback—because that's what this is—seems particularly ironic (and galling) when compared to efforts to improve moderation systems and reduce toxicity in the videogame realm. Riot Games, for instance, recently instituted a policy of holding Valorant and League of Legends players responsible for "off-platform" conduct; more notably, US senators have warned Valve that it may face "more intense scrutiny" from the government if it doesn't do something about the proliferation of extremist content on Steam. Yet Facebook, which is already not great in terms of spreading disinformation and toxicity, wants to be more like X, which is an absolute sewer, and that's apparently just fine.

The new Community Notes features on Facebook and Instagram will be phased in over the next couple months, beginning in the US.

Announced in a press release, the software is called G-Assist and it will be coming to the Nvidia App at some point in February. Just make sure you turn off the game filters if you download the Nvidia App, as we saw some performance issues with them turned on last month.

Outside of that, the Nvidia App, which was intended to replace GeForce Experience, is a solid bit of software that allows you to tune up your rig, easily download graphics drivers, and redeem codes for games when you buy new Nvidia gear.

G-Assist uses an on-device Small Language Model (SLM), which would explain why you need an RTX card and also means it can function, in theory, without needing a connection to the internet.

This tool can perform "real-time diagnostics and recommendations to alleviate system bottlenecks, improve power efficiency, optimize game settings, overclock your GPU, and much more."

As shown off in this short YouTube clip, it can even create graphs of CPU and GPU usage for monitoring purposes and to understand more about your rig. However, it's not all about graphs and figures and it can be used to explain Nvidia terms like how DLSS Frame Generation works or what specs are in your machine. It can even customize some basic features in Logitech G, Corsair, MSI, and Nanoleaf devices.

We had the chance to try out the app last June and came away pleasantly surprised. You can get the information it sources organically, if you know where to look, but this shines in allowing you to use casual and neutral language.

This could work in tandem with AMD's Adrenalin AI, which is designed specifically to answer AMD-based questions.

However, perhaps the most interesting part of the demo we tried isn't actually present in the press release. Labelled the "Knowledge Database", Nvidia's G-Assist was used in that demo last year to give advice to a player trying out Ark: Survival Ascended. It analysed what was on screen and gave recommendations to a player wondering how to spend their next skill point.

Razer has just announced its own gaming AI bot called Ava, and this seems to do something very similar, signalling how hard all these gaming companies are pushing to get their specific AI integration on the market.

Nvidia's gaming AI did not make an appearance in the most recent press release so we don't know if it's planned for the future or has been scrapped. Given the wide spread of information needed to give real-time advice in games, it seems very unlikely that version of G-Assist works on-device, and will likely be based on the cloud.

The version of G-Assist launching next month will primarily focus on optimising and finetuning RTX rigs, and it's one of the few uses of generative AI this CES I'm actually quite looking forward to. Hopefully, it lives up to those expectations.

"Imitation is the sincerest form of flattery, but Microsoft spoofing the Google homepage is another tactic in its long history of tricks to confuse users & limit choice," Tabriz writes. "New year; new low, Microsoft." This dunk is in response to a recent, and certainly underhanded, trick by Microsoft to redirect you to a faux-Google webpage if you search for "Google" on Bing, a sort of eerie facsimile of Google's layout, complete with Google-esque clipart.

I just tested it now, and it's still happening, although the "Microsoft Bing" cookies request at the bottom does somewhat ruin the kayfabe. As our own Joshua Wolens described it, "I'd bet a billion it'd hoodwink your less tech-literate family members and friends. You know, the kind of people who might find themselves doing a Bing search for 'Google'."

As for that "long history of tricks", Tabriz isn't entirely wrong. Just last year, PC Gamer's own Rich Stanton was blindsided when the sheer audacity of a pop-up advertising Bing during his browsing that made him think he had some sketchy malware installed. Then there's the Bing wallpaper app, which was allegedly doing the same, along with the recommended settings automatically swapping your default engine for its unused self. I'm sort of desensitised to it now, and I only use the search bar for files anyway, but in 2022 Microsoft made it your desktop's default engine, too. The Verge, whose article on the subject Tabriz is quoting, actually has an updated list of all the times Microsoft's tried to sneak it past you.

The funny thing is, Tabriz isn't even the only Google employee taking pot-shots at this thing. Product manager Jeffrey Jose writes "shameful" in a similar post, while product strategy manager Apran Das adds: "Nice jugaad", presumably sarcastically, given he's reposted Tabriz's statement (a jugaad is a word that describes a sort of patchwork or roundabout solution to a problem).

I mean, listen. I'm about as nervous about Google's techno-monopoly as any person with a passing knowledge of science fiction. The company has even been ordered to sell the very web browser in question after a judge ruled that it had acted illegally in order to maintain its iron grip on the market. But from a consumer perspective, you can't just 'here comes the aeroplane!' a search engine, Bing. We're grown adults. I only eat my broccoli if it's a spaceship, like any self-respecting member of society.

I've reached out to Microsoft to see if it'd like to explain this particular rugpull. I'll update this article if I receive a response. In the meantime, I'll just have to live with the fact that we live in a society just a few steps away from Google and Microsoft meeting in a parking lot to dance battle with each other.

Announced in a press release, the "intelligent streaming assistant" is made up of three central functions. The first is as a '3D Sidekick'. This AI functions as another host, which can answer questions you have in regards to the content you are reacting to, it can research topics for you, and it can even be used for "stimulating conversation when the stream is quiet by commenting on in-game events and chat activity".

You can also customise the personality of the 3D Sidekick to fit the tone of your stream and give you something to go back and forth with. We haven't yet seen this in action but it's hard to imagine this won't feel a little stilted in practice. This is made in tandem with "industry leaders like Nvidia and Inworld". Nvidia itself just announced a whole host of AI implementations in video games like a PUBG teammate who can respond to requests.

Nvidia's ACE AI suite is powering the sidekick, including "audio-to-face animations" as well as the contextual clue to attempt to understand what is on screen. Inworld then works on the cognitive engine, dealing with inputs and turning those inputs into understandable commands.

If the 3D Sidekick doesn't interest you, another function of this implementation allows the AI to function as a Producer behind the scenes, changing scenes dynamically to account for what's happening on screen.

A lot of live streamers will have jokes, gags, and references made as filters and scenes during their stream and this producer can activate them while you chat to your audience. As well as this, it can dynamically produce video and audio cues to add to your stream.

Finally, the AI can function as a 'technical assistant', troubleshooting problems you have with Streamlabs, from setting the app up to creating scenes. This is intended to make streaming for the first time more intuitive. The assistant can monitor streams for technical problems.

Though the first implementation of AI strikes me as a bit gimmicky long term, the latter two could help streamline the streaming process, especially for those who don't have the resources to hire a full team. We haven't seen this in action yet so we'll have to wait for its implementation "within the first half of 2025" to see it for ourselves.

Razer's "Project Ava" is one such AI implementation that, at first glance, boggles and baffles. That's because it's essentially a backseat gamer, and it seems Razer might have missed the memo that "backseat gamer" is usually a derogatory term.

But that's not entirely fair, because initial impressions aside, and as much as it pains me to say it about anything AI, Project Ava might have some genuine benefit for gamers. Project Ava, Razer says, can act as "the ultimate AI esports coach" and an "all-knowing AI game guide". It's essentially an AI assistant that monitors your gameplay and gives you vocalised hints in real time.

The idea is that it can pull from lots of different sources to give you real-time advice while gaming, whether that's in a single-player game like Black Myth: Wukong or an esports title you might need a hand with.

The bot, Razer says, offers "expert AI advice that draws from a wealth of community wisdom and knowledge bases. This allows gamers to stay fully immersed in the challenge without ever having to tab out. Gamers can bring up Ava while they are taking a quick break, hear Ava in their headset while they are in the action, or keep it simple with a chatbox. Whatever the choice, Ava is always game to help users win."

Project Ava also includes AI-aided single-click PC and game optimisation, but that's less interesting to me than all this backseat gaming malarkey. Malarkey, that is, which the more cynical might consider to be a form of cheating.

At the very least, Project Ava raises questions about where we want to draw the boundary between simple assistance and cheating. After all, one can have a build guide up to read while playing League of Legends, and we usually consider that to be fine. Does that assessment change if we have a bot vocally telling us what to buy and where to go?

I don't have an answer to that, but I imagine as more and more tools like this inevitably crop up, we'll be seeing ongoing debates about these things. (I should also note that although Razer demonstrated Project Ava with League of Legends, Razer says Project Ava isn't endorsed by Riot Games.)

It might, for instance, be most legitimately used while playing solo for practice. Thinking back to my Starcraft 2 days, I would certainly have benefitted from a real-time vocalisation of a chosen build order while playing, or even just a "probes and pylons" reminder. But I'm not sure how I'd feel about having access to that during a live 1v1.

There's also the fact that, while we found Project Ava worked well in Razer's demo, it does seem to pull game guide data straight from the net for reference. Which is fine, I suppose, as those guides are public, but it brings up the same old "Should AI be able to repackage human-made content?" debates we're all already used to.

It doesn't only pull from guide data, though. Razer says Ava considers pro player/coach insights, real-time event logs, item purchase history, public games played, public game guides/data, player gameplay style, player settings, and voice prompts. It works its AI magic on all that and spits out its in-game text, voice advice, or post-game analysis.

If this is a sign of things to come in 2025, AI might really have us in for a brave new world of PC gaming before too long. Let's just hope it comes across as more "helpful coach" than "backseat gamer" in practice.

Nvidia has announced a partnership with a handful of developers to create AI teammates, which is informed by Nvidia's own generative and responsive AI. This is to say that the AI teammates here are a little more intelligent than those you may have come across before.

AI that is used to dictate what NPCs do and what paths they take is generally fairly different to the AI used in creating images with six fingers and generating documents. However, Nvidia is combining the two, alongside PUBG publisher Krafton to make PUBG Ally.

This new AI system allows you to chat with an AI-led teammate, who can take orders and fire at enemies. In the YouTube clip (via The Verge) showing it off, the player asks for some ammo and a better piece of armour, and the bot not only finds it and pings it, but says it has done so. So far, so good. It is already more communicative than the average player I come across in PUBG.

As well as this, the AI teammate spotted an enemy, let the player know, and laid down covering fire. Of course, this is just a proof of concept and highly scripted so we don't know how it will shape out in actual games but this could be a genuinely good use for AI in games.

Just last year, Ubisoft announced the potential use of generative AI for NPCs in singleplayer games and I've always thought this feels like a gimmick, replacing the bespoke experience of talking to scripted NPCs with something I'd get bored of very quickly. However, being able to dynamically communicate with a bot could be good for when my friends are offline and I just want one more game. And adding a multiplayer chatbot doesn't run the risk of pushing artists out of games like other uses of generative AI.

However, we don't quite know how that will affect ranked mode and similar competitive play as having a teammate who is literally designed by AI to do everything you want it to could result in complaints of cheating. In the blog post for this new tech, Nvidia announced it would also be used for battle royale game Naraka: Bladepoint, AI characters in Sims-like Inzoi, and as smart enemies in MIR5.

Nvidia ACE, first announced at Computex 2023, is also powering conversational AI, like that in the Ubisoft post earlier on in the murder mystery game Dead Meat, AI People, a sandbox game, and ZooPunk, the follow-up to F.I.S.T: Forged in Shadow Torch.

Generative AI is worming its way into many facets of gaming and, though I rather dislike the prospect of AI-generated dialogue and art, slapping a chatbot onto multiplayer bots feels like it could work with some finetuning. Though Bladepoint's AI companions are set to launch in March this year, we don't yet have word on the rest.

We'll need to test out this tech for ourselves to see if its gets close to the chaos of playing with a few buddies.

Over the last few years, we have seen countless companies cash in on the popularity of AI with their own chatbots, image-generation tools, and more. Creatively named AMD Adrenalin AI, this is only possible in the upcoming line of AI-capable graphics cards from AMD, implying a level of on-device AI use, though we don't know to what extent that is the case.

We also don't yet know how this AI is being done and through which AI model. It could be in-house, which would explain part of its relatively small use case, or adapted from a larger model, which could make sense as there's a full-on image generation tool there.

The ability to summarize local documents means it can effectively go into PDFs, take out the necessary information, and give that to you. It could also be a good tool for clearing out storage as that's one of the bigger places for bloat as devices age.

However, we don't yet know if it can sum up multiple files at a time, or entire documents, so we will need to wait for this information before we figure out how useful it could be as a cleanup tool.

Finally, you can ask it specific AMD-related questions, like what AMD's frame generation technology is, what RDNA is, etc. This is effectively a chatbot designed to only answer specific, fairly niche questions on your rig.

Once again, we don't know the perimeters of this or what happens if you dare to stray outside of questions about AMD but, like any AI implementation, we'll have to get hands-on to see how this all works.

Interestingly, AMD's RDNA 4 architecture has specific improvements for AI and its upcoming range of AI chips for mobile use in the likes of the ASUS ROG Flow Z13 is built with Copilot+, Microsoft's AI suite in mind. This means we don't really know to what level users are actually expected to use AMD Adrenalin AI.

The next wave of graphics cards, as well as AMD's AI, is set to launch in Q1, 2025, so we should know a lot more about it in the coming months.

Most people aren't me, though, and quite sensibly use browsers that surpass 'basically fine' and search engines that do better than 'worthless', which is presumably why Microsoft seems to be trying to trick Bing users into thinking they're actually on Google. As spotted by Windows Latest, if you open up a Bing tab right now and use it to search 'Google', the search engine will return a big honking box that looks suspiciously like Google on top of the regular search results. Here's an example to save you the psychic damage of using Bing yourself.

Which seems downright egregious if you ask me. OK, sure, it could be worse: Microsoft could have wiped out the normal search results entirely, or worked some algorithm magic to make full-fat Google something other than the top search result, but it's still pretty bad. Not only does the page bear a striking resemblance to everyone's default search engine (featuring some kind of doodle), but Bing actually subtly scrolls you down when it loads the page to obscure the big ol' Microsoft Bing banner at the top.

It probably wouldn't fool you or me, but I'd bet a billion it'd hoodwink your less tech-literate family members and friends. You know, the kind of people who might find themselves doing a Bing search for 'Google'. It's frankly pretty gross, and seems like a tacit admission that, yeah, no one wants to use Bing, so the best it can do is dress itself up as the thing people actually want instead. It's like if Pepsi started distributing itself in red cans.

Plus, I mean, imagine being bad enough that disguising yourself as Google is a step-up? It might be the default search engine for most of us, but in an era of AI garbage and SEO-massaged slop pumped out into disposable website after disposable website, even the king of search is looking pretty sickly these days. Maybe we need to get back to looking things up in the library.

Microsoft casually dropped the monstrously huge sum of money in a long-winded blog by Brad Smith, Vice Chair & President of Microsoft, titled 'The Golden Opportunity for American AI' (via The Register). It starts by making it clear that AI is very much the number one focus right now, and that advances in AI technology and infrastructure need to continue. Microsoft also wants to promote training programs to get more people working in the field and it also wants America, as a whole, to export its AI systems to "allies and friends."

To ensure all this happens, "Microsoft is on track to invest approximately $80 billion to build out AI-enabled datacenters to train AI models and deploy AI and cloud-based applications around the world."

That kind of money makes the $10 billion it invested in OpenAI, back in 2023, seem like chump change. It's equivalent to almost all of Microsoft's net income in 2024. I know ridiculously large sums of money and AI just seem to be the norm these days, but $80 billion really is bonkers.

No company would spend that kind of money, though, if it didn't have a clear plan on how to recoup the investment and make a decent return from it. So what can we expect to see from Microsoft in this regard?

In its most recent financial statement, CEO Satya Nadella notes that "[our] growth depends on securely delivering continuous innovation and advancing our leading productivity and collaboration tools and services, including Microsoft 365, LinkedIn, and Dynamics 365" and that "digital transformation and adoption of AI continues to revolutionize more business workstreams for organizations in every sector across the globe."

In other words, since AI is already front-and-centre in every software package that Microsoft makes right now, one should expect to see it being promoted even more so. If you were hoping that the lacklustre sales of AI-ready PCs would mean the end of Copilot and Recall, then you're going to be disappointed.

The blog also talks about how Microsoft expects AI to eventually become another GPT or general-purpose technology because AI will apparently just be like electricity, ironworking, and computer chips in the same way they've "boost[ed] innovation and productivity across the economy." To me, that sounds like Microsoft wants its Azure AI service to be ubiquitous in the business world, with every company tapping into its machine learning subscription services.

PC gaming has, so far, avoided the full onslaught of AI, LLMs, and whatnot, but I wouldn't be surprised if 2025 is the year that it really kicks off. It'll be interesting to see what AI stuff gets mentioned at CES this week because it will give us a clue as to how bad (or good, let's be fair here) things might become.

On an average day about a dozen new games are released on Steam. And while we think that's a good thing, it can be understandably hard to keep up with. Potentially exciting gems are sure to be lost in the deluge of new things to play unless you sort through every single game that is released on Steam. So that’s exactly what we’ve done. If nothing catches your fancy this week, we've gathered the best PC games you can play right now and a running list of the 2024 games that are launching this year.

Sterling Shroud

Steam ‌page‌ ‌
Release:‌ January 4
Developer:‌ Squid Snooze Studio

What better way to kick off another year of this column than with the nostalgic austerity of a monochrome point 'n' click adventure? Sterling Shroud is a horror game that kinda looks like an old Apple IIE obscurity: thin white lines etch out rough approximations of real places, and it's up to your imagination to fill in the rest. The setting is a camping trip gone wrong, and the format is the usual blend of screen-by-screen first-person navigation and puzzles aplenty. I love the art style.

Puck

Steam‌ ‌page‌
Release:‌ January 5
Developer:‌ Nils Asejevs

Given the recent success of stripped back multiplayer experiences like Webfishing and STRAFTAT, I wouldn't be surprised to see Puck—a no BS online hockey game—find its audience. And by no BS, I mean there aren't even really any rules: all you have to do is get the ball in the goal; there's no "off side" or fouls or anything like that (is my complete lack of hockey knowledge showing here??). In other words, this is probably a fun thing to jump into for ten minutes now and then. You might as well too, because it's free.

Subterror

Steam‌ ‌page‌ ‌
Release:‌ January 4
Developers:‌ Team Monumental

After a stint in early access, this cooperative horror game set at the bottom of the ocean is now feature complete. Teams of up to four players will need to escape abandoned underwater biodomes playing host to a variety of environments and wildlife. The domes are procedurally generated and range a bunch of biomes including deserts, forests and tundras. It looks like a richly atmospheric affair—kinda like a more stylistically varied System Shock 2—and while coop appears to be the best way to play, you can go in solo as well.

Beyond Citadel

Steam‌ ‌page‌ ‌
Release:‌ January 3
Developer:‌ doekkuramori

This is the sequel to 2020 anime FPS The Citadel, and while that earlier game looked like Blake Stone remade in the Doom engine, Beyond Citadel is a generational leap into the Quake era (though it retains 2D sprite enemies). While it doesn't seem to offer much new to the now-well established boomer shooter genre, it does have atmosphere to spare and, probably most importantly for its audience, an anime aesthetic unusually under-utilised in the FPS space.

Wood and Flesh: A Candleforth Short Story

Steam‌ ‌page‌ ‌
Release:‌ January 5
Developer:‌ Under The Bed Games, Bibiki

This breezy point 'n' click adventure is a bite-sized follow-up to last year's Tales From Candleforth, which was a surreal puzzle-oriented take on twisted fairy tales. This "short story" is basically a sample of a forthcoming sequel, and takes the series' "cozy horror" vibe and wraps it around another bunch of puzzles and escape room scenarios. This is free, as these "prologue" samples tend to be, so it might be worth some time if you're in the mood for oddball horror.

The image was posted by Dittmann as part of a rambling defense of Musk's stewardship of X, his ability to appropriately self-medicate, and perhaps most famously, that "Elon is a father who gets lots of sex." The messages have since been deleted but are preserved for posterity (and, I must say, irony) in various posts on X. 

(There's also a 4chan archive if you want to go deep on the whole thing, but do note that it's awash in racism, homophobia, and transphobia—it's 4chan, after all—and adjust your expectations accordingly.)

What really got attention, though, is a shot of the X interface posted by Dittmann that seemingly revealed he has access to admin privileges on the platform. The image of his interface includes links to account switching, the admin portal, and bans—options not available to regular X users. 

That in itself is very odd indeed, but it carries extra weight because it comes on top of earlier allegations that Musk and Dittmann are one and the same: A July 2023 report by The Independent, for instance, said that Dittmann's "voice and laugh sounds remarkably similar to Mr Musk's," a point others on X have noted independently. Dittmann has also apparently referred to Musk in the first person on at least one occasion, and Musk's own daughter, Vivian Jenna Wilson, (via The Verge) also seems to believe that they're the same person.

Still, the image posted to 4chan isn't conclusive evidence by itself. A source at X told The Verge that the actual interface used by X employees doesn't look like the one seen in the image, implying that it was faked in order to encourage the rumor about Dittmann being a sockpuppet account. It's not even certain that Dittmann, whoever he is, posted it: 4chan users can use pretty much whatever name they want when starting new threads, so this could all simply be a bored third party looking to crank people up for fun.

But on the other other hand, Musk has done exactly this sort of thing before. In a 2024 lawsuit deposition reported by the Huffington Post, Musk copped to using sockpuppet accounts, including one in which he pretended to be a three-year-old child—although he said that account was only used "for testing."

Is Musk Dittmann? Is Dittmann Musk? What makes all of this so exhaustingly stupid is that it's impossible to say. In a sane world, claims that the wealthiest man on the planet is using a long-term sockpuppet to prop himself up with ego-boosters that sound like they're written by a pre-teen would be dismissed out of hand as unmitigated nonsense probably being shared by those same delinquent pre-teens. But we don't have that luxury in our world. We have speculation, anonymous denials, and—worst of all—the actual possibility that, yeah, Musk is Dittmann.

It's impossible to prove a negative and so there's a good likelihood that this madness will drag on for a long time to come, unless Musk either biffs it so badly that denial isn't possible, or he's forced to reveal the truth in another lawsuit deposition. For now, we can at least take some small comfort in the one thing in the online world that never lets us down: memes.

Update: Following the initial kerfuffle over Adrian Dittman's real identity, which blew up even further after Dittmann took part in a three-hour stream with Twitch streamer ConnorEatsPants, Maia Arson Crimew and Ryan Fae published a deep-dive analysis which they said provided "concrete evidence" that Dittman is not Elon Musk. Or, at least, that Adrian Dittman is a real person and not merely a construct of Musk's imagination.

It's a convincing argument as far as it goes, but leaves open the question of whether Musk has access to or control over Dittman's social media presence to any extent, which is a much more difficult question to answer. Musk himself seemed eager to continue fanning the flames of confusion, writing on X over the weekend (presumably not as a serious confession, but who can tell), "I am Adrian Dittmann. It's time the world knew."

I think it's still safe to say that for now, the world does not really know at all.

Since it was made open source, the 1992 shooter has been designated a veritable "can you run it" of potato hardware, finding its way onto teletext, volumetric displays, and even inside Doom itself (so you can gib while you gib).

So why nobody thought to make it a captcha until now boggles my mind, but I'm glad it's finally been done. We had a Doom-themed one before, but not one that was actually Doom.

The Doom captcha is a WebAssembly app, which runs low-level code in-browser, and it runs mighty fine based on my testing.

Testing which may or may not have confirmed I'm an awful gamer, as I failed over and over again to kill the three enemies required to "prove I'm human". I guess all those calling me a bot in Counter-Strike had a point.

It's a traditional arrows-and-spacebar affair, too. So be prepared to hearken back to CRT-era muscle memory—if such memory still sits in your brain bank—if you want to see those pixels spatter. Oh, and if you want to prove you're a human being.

The only question is whether and where this captcha will be used. I'd personally like to see donned on some very serious, official-looking websites. Government ones, perhaps? There's so much that could be done. "Rip and tear into the online portal", etc. etc. Hop on it, webdevs.

NTDev shared on X that they've managed to get Windows 11 running on less than 200 MB of RAM. They say: "Somehow I managed to make Windows 11 (through tiny11 and all that stuff, of course) run on just 184MB of RAM, which even though it's in Safe Mode I still think it's absolutely insane, and possibly a record."

Tiny11 is NTDev's application which lets you run Windows 11 on underpowered machines by getting rid of some of the pre-installed bloatware such as Camera, OneDrive, and so on.

Although it's not officially supported, recent Tiny11 builds should allow you to receive Windows updates—though nothing's guaranteed. And some of what it gets rid of can cause problems, if you're not careful. For instance, not having Edge installed could potentially cause problems with some widgets that rely on running Edge scripts.

Such Edge scripts might be especially important for the latest Windows AI features, too, given Copilot essentially uses embedded Edge scripts via WebView2. Then again, I'm sure some users would consider getting rid of (or breaking) these AI features a benefit of the modified install.

And call me a cynic, but the latest Windows install seems pretty, as they say, "borked" regardless, so perhaps slimming everything down will actually help in this regard.

184 MB is frighteningly little RAM, though, and I can't help but wonder just how usable Windows is when using such a low amount of memory. Certainly not browsing, as Task Manager is currently showing me about 5 GB being used on Chrome alone right now.

At the very least, Notepad seems to work with such low memory, as NTDev's screenshot attests. Perhaps a minimal install like this could be used as for note-taking, then. It's almost making me want to shovel out my old Asus Eee PC (yes, I really do still own one) and try it out. Watch this space, I guess.

I may not have mentioned this yet (I'm pretty sure I mentioned it in one of my first PC Gamer articles) but I have a law degree. That has made me love the minutia of the legal system but also keenly aware of how complicated and drawn out it can all get. Though the United States' enforcement of anti-trust principles as of late is likely better for the consumer and broader market, the latest hearing isn't the end of the case.

United States v. Google LLC (1:20-cv-03010) began proceedings in October 2020 and was decided in August of this year. This is a different case from the confusingly titled United States v. Google LLC (1:23-cv-00108), which started in 2023, and had its closing arguments in November 2024.

The latter case is specifically about Google's control and alleged monopoly over its advertising, whereas the former is about its search engine. Through deals with companies like Apple to be the default search engine on the iPhone and Google's ownership of the popular mobile OS Android, Google has become the search engine giant it is today.

There's a reason why it's a popular joke to jokingly insult the 'five users of Bing' or how behind the market Microsoft's Edge is. That browser is known as the app you use to get Chrome, then never touch it again. In fact, Windows periodically sends reminders that 'you can just use Edge, you know?' and yet many don't. I write this typing on Google Docs, through Google Chrome, so maybe I'm part of the problem.

Fundamentally, most browsers aren't too dissimilar from each other—yet Chrome holds almost 70% of the market share on a consistent basis. This lawsuit took aim at the browser, and after years of deliberation, the DOJ has proposed that Google be forced to sell Chrome, alongside a few other efforts to stop Google from further monopolising the market.

Judge Amit Mehta, who presides over the case, has ruled that Google acted illegally to develop and maintain its monopoly. As well as advising that Google sell Chrome, the DOJ proposed it be barred from reentering agreements like those that made it the default web browser, whilst also recommending it share its data with rivals.

It's also been recommended that Google either sell off Android or avoid making Google the default option on Android devices. The DOJ makes these recommendations to Judge Mehta, where Judge Mehta can mediate and come to an agreement.

This all seems like great advice to stop a company that has acted in a way that the court deems to be monopolistic. Controlling such a huge part of the market not only discourages others from entering but can push those in the market out. Once a company has a big enough control over a resource, like all the digital real estate Chrome has, it can essentially make its own rules.

Outside of court orders like this, there aren't ways to hold huge corporate entities accountable when they are acting against the interests of their users. If Chrome, Edge, Opera, and more were equally viable choices, you could afford to swap from one to the other if you didn't like any decisions made.

Of course, selling Chrome won't shrink its user base overnight. Though the other actions recommended by the DOJ would help with that too, it's important to note that the litigation against Google is more than just a singular case, it's one of multiple antitrust suits levelled against the company in the last few years.

The worry isn't specifically and exclusively about Chrome, but all the actions made to keep it where it is today. Google as a company doesn't appear to be trusted to handle something like Chrome in legal ways.

Published on November 21, just a single day after the DOJ proposed Google sell Chrome, Kent Walker, the global affairs and chief legal officers of Google and Alphabet (Google's parent company), took to the Google Blog to declare that this decision "would hurt consumers".This piece makes the argument that the proposal would endanger the security of users by undermining the quality of the services and disclosing Google's research to "foreign and domestic companies".

It also argued that Google's investment into AI would be chilled by this decision, effectively warning that Google's own advancement of AI would be held back as a result.

The blog also argues that it would not only hurt casual consumers' access to Google search but would hurt companies reliant on Google, like Firefox. Finally, it argues this is a slippery slope and an overreach of the government on "your online experience".

None of these arguments, on their face, are necessarily wrong—but I'd argue they're lacking a little in nuance. Selling off such a huge entity could indeed leave consumers open to bad practices, both technically and ethically, from the potential buyer. It is also true that Google's research into the likes of Google Gemini would make less sense without owning the broader ecosystem. Also, with so many browsers being based on Chromium, Google's open-source web browsing project, there's a chance for security or quality problems as a result.

However, that's also part of what the current process in the lawsuit is for. Judge Mehta has argued Google acted illegally and Google does have the chance to appeal, which it has been suggested it will do. If it fails to file as such, a further trial will be held in April next year to come up with an answer to remedy the alleged wrongdoing.

In this case, Google's arguments will be heard and an answer will be found. Many of the arguments brought forward by Google could be remedied through further litigation and aren't themselves arguments against the suit. It's important to remember that the DOJ's recommendation isn't law, it's merely a suggestion based on the facts of the case, and other remedies can be found.

Though we have many arguments and a decision, this case is anything but final and it's hard to have too concrete an understanding until a later date. Though the breaking up of parts of Google seems to have been necessary for some time as argued by Judge Mehta, the specifics of how that will happen will dictate how successful the potential selling would be.

Chrome is a major part of much of how Google currently operates and measures will need to be taken to not lose many fundamental parts of the internet seemingly overnight. A heavy-handed approach was needed for much of the court case to even bring this suit against Google but a finer touch is needed for those specifics. Though we won't understand how fine that is for some time.

Microsoft has been logging many of the various known issues on its Windows release health page, and the most recent entry relates to Auto HDR. Some users have been experiencing incorrect game colours with "certain display configurations" and even full-on crashes in some games.

Microsoft has applied a compatibility hold on devices with Auto HDR enabled, meaning that these devices will not be offered the option to install 24H2 via the Windows Update release channel if they haven't installed it already.

However, that's far from the only 24H2 bug doing the rounds this holiday season. Second on the list is an entry mentioning audio issues with a "limited set of devices from one manufacturer" namely devices using Dirac Audio with cridspapo.dll.

Again, Microsoft says it's put a compatibility hold on affected PCs—but both myself and my colleague, Nick Evanson, have had audio device issues since installing the update, and neither of us uses Dirac Audio. In my case I've had audio devices disappear, refuse to switch over (which is fun for meetings!) and odd crackling.

All of this is directly post-update, which for me also broke many of my drivers and required me to reinstall them manually. Oh, and my Windows notifications keep disappearing and reappearing en masse, too.

Polling the hardware team for odd PC experiences since installing the update reveals yet more strange behaviour. Our Jacob Fox has experienced animation glitching when minimising and maximising individual windows, which a quick Google search reveals seems to be an ongoing issue for many users.

He's disabled animations entirely to fix it, so spare a thought for poor Mr Fox this holiday season with his barebones Windows experience.

Nick has ongoing issues with Chrome, which (directly after the installation of 24H2) occasionally freezes page rendering halfway through, leaving the bottom half stuck in place but the top half still active. And as for myself, it's not just audio and notification issues that have me scratching my head.

I noticed odd stuttering and hangs in games that ran smoothly pre-update, including Diablo 4 and No Mans Sky, both of which usually run consistently at over 100 fps on my machine. Spending an evening scouring forums for a fix (not exactly my favourite way to spend a night off), I noticed that some people were recommending turning off fast startup and disabling hibernation to solve the issue in certain games.

And wouldn't you know it, that's worked for me too. These are anecdotal experiences of course, but I'd say it was worth a try if you've experienced odd hitching and micro stutter issues since installing the update yourself.

All this, and I haven't even got to the Ubisoft debacle, in which 24H2 was prevented from being sent out to machines with certain Ubisoft games installed on them, like Star Wars Outlaws and Assassin's Creed: Valhalla, in order to prevent crashing. Ubisoft has been releasing hotfixes to solve the problem, but certain games are still yet to receive a fix.

Previous issues have included the 24H2 audio jump scare bug, the Western Digital SSD BSOD bug, and a borknado of other pre-public launch problems.

So, it seems 24H2 continues to cause headaches for many. I can't remember a time a major Windows update didn't cause widespread issues, but even so. this latest patch seems to cause more than most.

So here's hoping Microsoft squashes these bugs sooner rather than later. I have games to play over the holidays, you see, and I'd rather not spend my time off scouring the forums for yet more hackaround fixes.

Today, for day 10 of its "12 days of OpenAI" Christmas lead-up event, OpenAI has gone further than I expected by unleashing its chatbot from the tethers of the internet and into the telecom airwaves. Yes, you can now call ChatGPT or WhatsApp message it. Unfortunately (or fortunately?) you can only do the former if you live in the US, but if you live elsewhere you can still do the latter.

The number is 1-800-CHATGPT, and yes, it's toll-free. Which is how OpenAI is marketing it, by the way: "Call toll free, 1-800-ChatGPT." Nice. That's 1-800-2428478 for those unfamiliar with traditional keypad texting.

OpenAI's Keven Weil says that this is all about "making it [AI] as accessible as possible to as many people as we can."

As you can see from the image below, I found ChatGPT WhatsApp to work just fine. The QR code I've slapped on to the right-hand side is taken from OpenAI's video. If you scan it, it opens up the WhatsApp app and takes you to a conversation with ChatGPT.

I can't test the phone calls as I'm not based in the US, but judging from the phone calls the OpenAI reps make in the video, that seems to work just fine.

I'm slightly jealous of those of you living across the pond after watching that video, too. If I could get over the initial embarrassment of doing so, I reckon being able to speed-dial ChatGPT could come in quite useful. A bit like using Siri, but... well, more conversational and better (sorry, Apple). Plus, you could even do so on a dumbphone.

Just don't go using it as replacement for, you know, actual human contact. It's not a conscious human being. Maybe try hanging up on it a few times to get your mind used to that fact. It's not rude, it doesn't feel anything. Don't believe me? It even says so, unprompted, in our chat above.

So there you go, that's day 10 from OpenAI for you. There've been some smaller announcements from OpenAI over this 12-day bonanza, but some bigger ones, too. On day three, for instance, the AI company unleashed Sora, its video generation model. And now we have this: ChatGPT phone calls and ChatGPT on WhatsApp.

I wonder what the last couple of days will bring. I'm not sure much could top this, to be honest. Well, not much that doesn't have me breaking open the Skynet Defence Manual. And that wouldn't be very Christmassy, would it?

As reported by the Financial Times, the tech consultancy Omdia estimates that Nvidia's biggest purchaser of Hopper chips in 2024 was Microsoft, who bought 485,000 of them, this being over twice as many as any other company. Meta, for example, bought 224,000 of them, Amazon bought 196,000, and Google bought 'just' 169,000.

Surprisingly, though, two of Nvidia's biggest customers were Chinese ones, these being ByteDance (of TikTok fame) and Tencent (of numerous videogames fame). According to Omdia, these each ordered about 230,000 Hopper chips. These won't have been the most powerful ones Nvidia has at its disposal, though, given US-China export restrictions.

In other words, despite export controls, Chinese companies received more Nvidia chips than companies such as Meta, Amazon, and Google this year.

Apart from all the China biz, there are two other things that strike me about these numbers. First, and very simply: Holy moly does Nvidia churn out a lot of a AI chips. Second: Holy moly does Microsoft buy a lot of them.

Nvidia's part kind of goes without saying. The company is firmly cemented as the king of the ever-expanding AI infrastructure castle. So much so, in fact, that the company's CEO, Jen-Hsun Huang, has the chutzpah to claim that "just like we generate electricity, we're now going to be generating AI" in "AI factories".

As far as Microsoft's concerned, while I'm a little surprised to find the company more than doubling the purchases of any other, it also makes sense, especially given the partnership with OpenAI.

The AI industry can seem a little confusing when you start to look into how all these different companies relate to each other. But we shouldn't forget that while OpenAI technically isn't the biggest company in the AI space, most of the companies that are bigger actually rely on OpenAI's software and models and have partnerships with the company. So much is true for now, at least—although newer entrants such as Anthropic and Musk's xAI could make inroads.

And if we're talking OpenAI partnerships, Microsoft's is the one. To date, Microsoft has apparently invested $13 billion into OpenAI and is the exclusive provider of the company's cloud computing services. This partnership grants Microsoft all kinds of benefits, such as OpenAI model integration with Bing, Microsoft 365, Copilot, and so on, not to mention the ability to rent OpenAI-clad Azure servers out to customers for private or bespoke AI research or services.

Oh, and there's the simple matter of monetary ROI. But it's surely crude to speak of such things (profit motives in such a civil society? I think not).

People want AI and OpenAI is the biggest software-level solution, so Microsoft heavily invests in and partners with OpenAI, and Nvidia sells Microsoft a sweet, sweet stack of silicone to get the job done. Simple.

But who's the real winner? The end-user, of course!

Just kidding, the real winner's Nvidia, of course and as always. As Baron Harkonnen of Frank Herbert's Dune tells us: "He who controls the spice controls the universe." And Hopper's the spiciest spice in town, right now.

Well, it'll be Blackwell, soon, but that's Nvidia, too. Huang probably made the right choice going into tech and ditching a promising table-tennis career.

We all know that AI involves billions of dollars. One can't build data centres full of thousands of massive GPUs, pay the energy bills, and hire staff to program them for a mere pittance. But should that expenditure be used to create something for non-profit making or to create something that will ultimately rake in huge sums of money? When it comes to OpenAI, creators of ChatGPT, and Elon Musk, owner of xAI and Grok, it would seem they've not seen eye-to-eye on this matter.

OpenAI started as a non-profit AI research organisation in December 2015, before launching a for-profit subsidiary (OpenAI Global) four years later. The company was co-chaired by Sam Altman and Elon Musk until 2018 when Musk resigned from the board of directors, to avoid a conflict of interest with Tesla, which uses AI for its autonomous driving systems.

Over the years, other staff left OpenAI to start up AI companies of their own, and OpenAI Global attracted the attention of Microsoft, which pumped billions of dollars into the subsidiary, securing a nearly 50% share of the company.

Earlier this year, however, Elon Musk sued OpenAI and Sam Altman for breach of contract, violating fiduciary duty, and unfair business practices—in short, because OpenAI had decided to target profits and commercial interests, over the company's original goal of pursuing AI for the good of mankind.

Fast forward to a few days ago and it would seem that all is not quite as simple as it sounds, because OpenAI has published redacted DMs and emails that show Musk wanted OpenAI to go for-profit in the summer of 2017, before then going on to "demand majority equity, absolute control, and to be CEO of the for-profit."

In one email supposedly from Musk, he wrote that "the Preferred A investment round (supermajority me) should have the right to appoint four (not three) seats. I would not expect to appoint them immediately, but, like I said I would unequivocally have initial control of the company, but this will change quickly."

While it's not an outright demand for the control of OpenAI, I think it's fair to say that the message doesn't read particularly well and the board rejected Musk's option. It tried to placate him, though, saying "We really want to work with you. We believe that if we join forces, our chance of success in the mission is the greatest. Our upside is the highest. There is no doubt about that. Our desire to work with you is so great that we are happy to give up on the equity, personal control, make ourselves easily firable — whatever it takes to work with you."

Not that this had any effect because Musk's response was unequivocal: "Guys, I’ve had enough. This is the final straw. Either go do something on your own or continue with OpenAI as a non-profit. I will no longer fund OpenAI until you have made a firm commitment to stay or I’m just being a fool who is essentially providing free funding for you to create a startup. Discussions are over."

From the various messages being now made public, Musk was seemingly never hugely opposed to opening up a for-profit OpenAI offshoot or even necessarily for the whole company to go that way. The issue seems to be how it would all be organised and funded. Which does make his current lawsuit look maybe a bit spurious: why is there an issue with a for-profit OpenAI now?

Where things go now isn't clear, though I imagine that Elon Musk will have a few things to say on the matter. As far as I can tell, there's no mention of OpenAI's claims on his X account but the man posts a lot so there could be something buried away. The sensible thing to do would be to leave this all in the hands of legal beagles, but I guess OpenAI's post rather sidesteps that.

Time to get some more popcorn in, methinks.