Big players, including Microsoft, with Copilot, Google, with Gemini, and OpenAI, with GPT-4o, are making AI chatbot technology previously restricted to test labs more accessible to the general public.How do these large language model (LLM) programs work? OpenAIs GPT-3 told us that AI uses a series of autocomplete-like programs to learn language and that these programs analyze the statistical properties of the language to make educated guesses based on the words youve typed previously. Or, in the words of James Vincent, a human person: These AI tools are vast autocomplete systems, trained to predict which word follows the next in any given sentence. As such, they have no hard-coded database of facts to draw on just the ability to write plausible-sounding statements. This means they have a tendency to present false information as truth since whether a given sentence sounds plausible does not guarantee its factuality.But there are so many more pieces to the AI landscape that are coming into play (and so many name changes remember when we were talking about Bing and Bard before those tools were rebranded?), but you can be sure to see it all unfold here on The Verge.HighlightsToday, 2:49 AM UTCRichard LawlerOpenAI CEO Sam Altman on DeepSeek R1: an impressive model.Now that a Chinese startup has captured a lot of the AI buzz, what happens next?The ChatGPT boss says of his company, we will obviously deliver much better models and also its legit invigorating to have a new competitor, then, naturally, turns the conversation to AGI.Screenshot: @sama (X)Jan 27Richard LawlerDeepSeek says its newest AI model, Janus-Pro can outperform Stable Diffusion and DALL-E 3.Already riding a wave of hype over its R1 reasoning AI that is atop the app store charts and shifting the stock market, Chinese startup DeepSeek has released another new open-source AI model: Janus-Pro.Input image analysis is limited to 384x384 resolution, but the company says the largest version, Janus-Pro-7b, beat comparable models on two AI benchmark tests.Correction: As TechCrunch notes, Janus-Pro image input is listed as limited to low resolution, not its output.Image: DeepSeekJan 27Emma RothMeta AI will use its memory to provide better recommendationsIllustration by Nick Barclay / The VergeMeta is widely launching the ability for its AI chatbot to remember certain details about you, such as your dietary preferences or your interests, the company said in a blog post on Monday. It will then use your past conversations, in addition to details from Facebook and Instagram accounts, to provide more relevant recommendations.Meta first started rolling out a memory feature for its AI chatbot last year, but now it will be available across Facebook, Messenger, and WhatsApp on iOS and Android in the US and Canada. Though you can tell Meta AI to remember certain things, like that you love traveling, it will also pick up important details based on context.Read Article >Jan 27Jess WeatherbedDeepSeeks top-ranked AI app is restricting sign-ups due to malicious attacksImage: Cath Virginia / The VergeAfter surging to the top of Apples App Store charts in the US, DeepSeeks AI Assistant is now restricting new user sign-ups. According to an incident report page, registrations are being temporarily limited due to large-scale malicious attacks on DeepSeeks services, though its unclear how these limitations are being applied.Existing users can log in as usual, DeepSeek said in its update. Thanks for your understanding and support. An alert banner on the DeepSeek web sign-up page says that registration may be busy, rather than entirely restricted, however, and encourages users to wait and try again if their application is unsuccessful.Read Article >Jan 25Wes DavisOpenAI has added its o1 model to Canvas.In an X post announcing the change yesterday, the company also said that Canvas, its ChatGPT coding helper feature, now has the ability to render HTML and React code.OpenAI added that Canvas has rolled out to the ChatGPT desktop app for macOS.Jan 25Richard LawlerCharacter.ai responds to a wrongful death lawsuit aimed at its chatbots.Last fall, Megan Garcia sued Character.AI, its founders, and Google over the death by suicide of her 14-year-old son, who had chatted continuously with its bots, including just before his death. In December, the firm added safety measures aimed at teens and concerns over addiction.TechCrunch reports that Character.ai has filed a motion to dismiss the case, which you can read in full here.Screenshot: C.ai Motion to DismissJan 22David PierceGoogles Gemini is already winning the next-gen assistant warsIllustration: The VergeOne of the most important changes in Samsungs new phones is a simple one: when you long-press the side button on your phone, instead of activating Samsungs own Bixby assistant by default, youll get Google Gemini. This is probably a good thing. Bixby was never a very good virtual assistant Samsung originally built it primarily as a way to more simply navigate device settings, not to get information from the internet. It has gotten better since and can now do standard assistant things like performing visual searches and setting timers, but it never managed to catch up to the likes of Alexa, Google Assistant, and now, even Siri. So, if youre a Samsung user, this is good news! Your assistant is probably better now. (And if, for some unknown reason, you really do truly love Bixby, dont worry: theres still an app.)Read Article >Jan 18Wes DavisMicrosoft opens testing for Windows AI searchImage: The VergeMicrosoft is testing AI-powered Windows search in a new dev channel build for Windows 11 Insider testers. Announced in October, it uses semantic indexing to let users search for local files using more casual language. Like other Microsoft AI features, youll need a Copilot Plus PC to use it.The feature applies whether youre using search boxes in Settings, File Explorer, or the taskbar. And you dont need to be connected to the internet for it to work, thanks to the NPU chips on Copilot Plus computers. For now, AI search is limited to Windows settings and files with image and text formats that include JPEG, PNG, PDF, TXT, and XLS.Read Article >Jan 13Wes DavisRecording is hard, so let AI do it is a bad take.In this interview clip that 404 Media spotted, CEO Mikey Shulman, who runs bad AI song generator Suno, says that its not really enjoyable to make music now.Having lost countless nights to it, and considering my days in recording studios were some of the best of my life, Shulman seems to be either flatly lying or has no idea what hes talking about.Jan 13Wes DavisMicrosoft drops its GitHub Copilot Workspace waitlist.More developers can now access Microsofts AI coding assistance tool thats been on a waitlist since its debut in April last year, company CEO Satya Nadella announced in a LinkedIn post on Sunday.GitHub developers can go here to try it out.Satya Nadella on LinkedIn: There is no more waitlist for GitHub Copilot Workspacethe most advanced | 136 comments[linkedin]Jan 9Wes DavisMicrosoft is reverting its Bing AI image generator because of quality complaintsIllustration by Haein Jeong / The VergeMicrosoft is rolling back a model upgrade to its AI-powered Bing Image Creator, reports TechCrunch. The rollback came after weeks of complaints by users that the tool just didnt work as well after Microsoft upgraded to a new version of the DALL-E 3 model on December 18th.Microsoft declined to comment on its decision to roll things back or offer specifics on what may be causing the gap between users expectations and its output.Read Article >Las Vegas police release ChatGPT logs from the suspect in the Cybertruck explosionImage: LMVPDNearly a week after a New Years Day explosion in front of the Trump Hotel in Las Vegas, local law enforcement released more information about their investigation, including what they know so far about the role of generative AI in the incident.They confirmed that the suspect, an active duty soldier in the US Army named Matthew Livelsberger, had a possible manifesto saved on his phone, in addition to an email to a podcaster and other letters. They also showed video evidence of him preparing for the explosion by pouring fuel onto the truck while stopped before driving to the hotel. Hed also kept a log of supposed surveillance, although the officials said he did not have a criminal record and was not being surveilled or investigated.Read Article >Dec 21, 2024Wes DavisGemini can now tell when a PDF is on your phone screenIllustration: The VergeIn the latest version of the Files by Google app, summoning Gemini while looking at a PDF gives you the option to ask about the file, writes Android Police. Youll need to be a Gemini Advanced subscriber to use the feature though, according to Mishaal Rahman, who reported on Friday that it had started rolling out.If you have the feature, when you summon Gemini while looking at a PDF in the Files app, youll see an Ask about this PDF button appear. Tapping that lets you ask questions about the file, the same way you might ask ChatGPT about a PDF. Google first announced this screen-aware feature during its I/O developer conference in May.Read Article >Dec 17, 2024Wes DavisNvidias $249 dev kit promises cheap, small AI powerNvidia announced the latest in its Jetson Orin Nano AI computer line, the Jetson Orin Nano Super Developer Kit. Sort of like a Raspberry Pi but for powerful AI processing, the tiny $249 computer packs more of an AI processing punch than the kit did before for half the price. Its available to buy now.The Jetson Nano line has been a low-cost way for hobbyists and makers to power AI and robotics projects since its introduction in 2019. Nvidia says the Nano Supers neural processing is 70 percent higher, at 67 TOPS, than the 40 TOPS Nano. It also has 50 percent more memory bandwidth, at 102GB/s, which should speed up those operations.Read Article >Dec 17, 2024Wes DavisDexcom adds AI reports to its OTC glucose monitor.Dexcoms Stelo continuous glucose monitor (CGM) for those with Type 2 diabetes is starting to use generative AI to write weekly reports with more personalized tips, recommendations, and education related to diet, exercise, and sleep than the template previously used.CNBC:Stelos AI reports dont give users medical advice, though Dexcom has been using an AI framework from the U.S. Food and Drug Administration to help guide the features development, [Dexcom COO Jake] Leach said.Dexcom's over-the-counter glucose monitor now offers users an AI summary of how sleep, meals and more impact sugar levels[CNBC]Dec 16, 2024Jay PetersGoogles Whisk AI generator will remix the pictures you plug inAn AI-generated image I made in Whisk using Googles suggested images as prompts. Image: Google via WhiskGoogle has announced a new AI tool called Whisk that lets you generate images using other images as prompts instead of requiring a long text prompt.With Whisk, you can offer images to suggest what youd like as the subject, the scene, and the style of your AI-generated image, and you can prompt Whisk with multiple images for each of those three things. (If you want, you can fill in text prompts, too.) If you dont have images on hand, you can click a dice icon to have Google fill in some images for the prompts (though those images also appear to be AI-generated). You can also enter some text into a text box at the end of the process if you want to add extra detail about the image youre looking for, but its not required.Read Article >Dec 16, 2024Alex HeathGoogle says the next version of its Sora competitor is better at real-world physics.In a post announcing waitlist sign-ups for its Veo 2 video model, Google says the next version brings an improved understanding of real-world physics and the nuances of human movement and expression.OpenAIs Sora notably struggles with physics, so it will be interesting to compare the results of Veo 2 when we eventually get access.State-of-the-art video and image generation with Veo 2 and Imagen 3[Google]Dec 15, 2024Wes DavisInstagrams head says social media needs more context because of AIIllustration by Nick Barclay / The VergeIn a series of Threads posts this afternoon, Instagram head Adam Mosseri says users shouldnt trust images they see online because AI is clearly producing content thats easily mistaken for reality. Because of that, he says users should consider the source, and social platforms should help with that.Our role as internet platforms is to label content generated as AI as best we can, Mosseri writes, but he admits some content will be missed by those labels. Because of that, platforms must also provide context about who is sharing so users can decide how much to trust their content.Read Article >Dec 14, 2024Wes DavisListen yall, its a sabotage.Folks in the online AI research community are upset after the worlds biggest AI conference, NeurIPS, gave its prestigious Best Paper Award to, among others, a controversial former ByteDance intern named Keyu Tian, writes Wired.ByteDance allegedly dismissed Tian for sabotaging colleagues AI research and hoarded resources for his own work accusations detailed in an anonymous GitHub blog calling for the award to be revoked.Former ByteDance Intern Accused of Sabotage Among Winners of Prestigious AI Award[WIRED]Dec 13, 2024David PierceSearching for the first great AI appImage: Alex Parkin / The VergeChatGPT launched roughly two years and two weeks ago. Now, as we near the end of 2024, the AI race is... well, where is it, exactly? Its more competitive than ever, theres more money being poured into new models and products than ever, and its not at all clear when or even whether were going to get products that make it all worthwhile.On this episode of The Vergecast, we talk about a lot of different AI news, all along a single trend line: the tech industry trying desperately to build a killer app for AI. (Ideally, for them, also one that makes money.) The Verges Richard Lawler joins us as we discuss Google Gemini 2.0, Project Astra and Project Mariner, and everything else Google is doing to put AI in the products you already use every day. We also talk through the new Android XR announcement, and Googles renewed commitment to making headsets and smart glasses that work. Its all an AI story, no matter how you look at it.Read Article >Dec 12, 2024Emma RothGemini AI can now summarize whats in your Google Drive foldersIllustration by Alex Castro / The VergeGeminis integration into Google Drive is getting a little more useful. In addition to summarizing documents or answering questions about a project, the AI assistant can now generate summaries of everything inside a folder.With the feature, you can open a folder and select the new Summarize this folder button at the top of the page. Gemini will then give you a breakdown of the folders contents. As noted by Google, you can use Gemini to find specific files inside a folder, or ask questions about it, like What is the theme of this folder?Read Article >Dec 11, 2024Kylie RobisonGoogles AI enters its agentic eraProject Astra is supposed to be more useful than ever and it knows where you are. Image: GoogleI stepped into a room lined with bookshelves, stacked with ordinary programming and architecture texts. One shelf stood slightly askew, and behind it was a hidden room that had three TVs displaying famous artworks: Edvard Munchs The Scream, Georges Seurats Sunday Afternoon, and Hokusais The Great Wave off Kanagawa. Theres some interesting pieces of art here, said Bibo Xu, Google DeepMinds lead product manager for Project Astra. Is there one in particular that you would want to talk about?Project Astra, Googles prototype AI universal agent, responded smoothly. The Sunday Afternoon artwork was discussed previously, it replied. Was there a particular detail about it you wish to discuss, or were you interested in discussing The Scream?Read Article >Dec 11, 2024David PierceGoogle launched Gemini 2.0, its new AI model for practically everythingIllustration: The VergeGoogles latest AI model has a lot of work to do. Like every other company in the AI race, Google is frantically building AI into practically every product it owns, trying to build products other developers want to use, and racing to set up all the infrastructure to make those things possible without being so expensive it runs the company out of business. Meanwhile, Amazon, Microsoft, Anthropic, and OpenAI are pouring their own billions into pretty much the exact same set of problems. That may explain why Demis Hassabis, the CEO of Google DeepMind and the head of all the companys AI efforts, is so excited about how all-encompassing the new Gemini 2.0 model is. Google is releasing Gemini 2.0 on Wednesday, about 10 months after the company first launched 1.5. Its still in what Google calls an experimental preview, and only one version of the model the smaller, lower-end 2.0 Flash is being released. But Hassabis says its still a big day.Read Article >Dec 11, 2024Emma RothGoogle built an AI tool that can do research for youIllustration: The VergeGoogle has just revealed a new AI tool called Deep Research that lets you call upon its Gemini bot to scour the web for you and write a detailed report based on its findings.Deep Research is currently only available in English to Gemini Advanced subscribers. If you have access, you can ask Gemini to research a particular topic on your behalf, and the chatbot will create a multi-step research plan that you can either edit or approve. Google says Gemini will start its research by finding interesting pieces of information on the web and then performing related searches a process it repeats several times.Read Article >Dec 10, 2024Wes DavisYouTubes AI-powered dubbing is now available to many more creatorsIllustration by Alex Castro / The VergeYouTube says it has expanded its AI-powered auto-dubbing to hundreds of thousands of channels in the YouTube Partner Program that are focused on knowledge and information. YouTube says it will bring the feature to other types of content soon.What the dubs consist of depends on the language used in the original video. If it was in English to start, itll be translated into French, German, Hindi, Italian, Spanish, Indonesian, Japanese, and Portuguese. If the starting video was made in one of those languages, YouTube will only produce an English dub.Read Article >More Stories