When Elon Musk launched his own AI startup, xAI, he touted a key advantage over his competitors: access to the vast trove of data from his newly acquired social media platform Twitter. By implementing new API fees on the network he quickly renamed X, Musk locked out other AI companies, maintaining exclusive access for his own models. And he began using X’s millions of users to test the results.

Musk has been using this distribution channel since xAI launched its first version of the Grok large language model, adding features like trending story summaries and AI-generated questions on posts as well as releasing the Grok chatbot (initially) to X users exclusively. Now, a slew of new AI features is coming. Per the findings of reverse engineer Nima Owji, the platform appears to be developing AI-powered post enhancements, including a feature that lets Grok modify your tweets. The chatbot also appears to be adding location-based queries, letting users ask about things nearby, like grocery stores.

xAI’s takeover of the platform once known as Twitter is so unmistakable that even its branding has crept into X’s most visible real estate, with “xAI Grok” now commanding prominent placement...

Read the full story at The Verge.

If you were looking for your own personal AI supercomputer, Nvidia has you covered.

The chipmaker announced at CES it’s launching a personal AI supercomputer called Project Digits in May. The heart of Project Digits is the new GB10 Grace Blackwell Superchip, which packs enough processing power to run sophisticated AI models while being compact enough to fit on a desk and run from a standard power outlet (this kind of processing power used to require much larger, more power-hungry systems). This desktop-sized system can handle AI models with up to 200 billion parameters, and has a starting price of $3,000. The product itself looks a lot like a Mac Mini.

“AI will be mainstream in every application for every industry. With Project Digits, the Grace Blackwell Superchip comes to millions of developers,” Nvidia CEO Jensen Huang said in a press release. “Placing an AI supercomputer on the desks of every data scientist, AI researcher and student empowers them to engage and shape the age of AI.”

Each Project Digits system comes equipped with 128GB of unified, coherent memory (by comparison, a good laptop might...

Read the full story at The Verge.

For the last day of ship-mas, OpenAI previewed a new set of frontier “reasoning” models dubbed o3 and o3-mini. The Verge first reported that a new reasoning model would be coming during this event.

The company isn’t releasing these models today (and admits final results may evolve with more post-training). However, OpenAI is accepting applications from the research community to test these systems ahead of public release (which it has yet to set a date for). OpenAI launched o1 (codenamed Strawberry) in September and is jumping straight to o3, skipping o2 to avoid confusion (or trademark conflicts) with the British telecom company called O2.

The term reasoning has become a common buzzword in the AI industry lately, but it basically means the machine breaks down instructions into smaller tasks that can produce stronger outcomes. These models often show the work for how it got to an answer, rather than just giving a final answer without explanation.

According to the company, o3 surpasses previous performance records across the board. It beats its predecessor in coding tests (called SWE-Bench Verified) by 22.8 percent and outscores OpenAI’s Chief Scientist in competitive programming. The model nearly aced one of the hardest math competitions (called AIME 2024), missing one question, and achieved 87.7 percent on a benchmark for expert-level science problems (called GPQA Diamond). On the toughest math and reasoning challenges that usually stump AI, o3 solved 25.2 percent of problems (where no other model exceeds 2 percent).

OpenAI

The company also announced new research on deliberative alignment, which requires the AI model to process safety decisions step-by-step. So, instead of just giving yes/no rules to the AI model, this paradigm requires it to actively reason about whether a user’s request fits OpenAI’s safety policies. The company claims that when it tested this on o1, it was much better at following safety guidelines than previous models, including GPT-4.

For the 10th day of “ship-mas,” OpenAI rolled out a way to call ChatGPT for up to 15 minutes for free over the phone using 1-800-CHATGPT.

The feature was a project spun up just a few weeks ago, OpenAI’s chief product officer Kevin Weil said on the livestream. Users can now call ChatGPT in the US and message via WhatsApp globally at 1-800-242-8478. The 15-minute limit is per phone number per month, so really, you could spin up a few Google Voice numbers to get as much time with it as you want.

The phone number is built using OpenAI’s Realtime API, and the WhatsApp feature is powered by GPT-4o mini through an integration with the WhatsApp API.

OpenAI

OpenAI sees this feature as an important stepping stone for newcomers to AI, since the service represents a simplified version of ChatGPT compared to its web-based counterpart and offers a “low-cost way to try it out through familiar channels.” The company notes that existing users seeking more comprehensive features, higher usage limits, and personalization options should continue using their regular ChatGPT accounts through traditional channels.

Funnily enough, Google launched a similar tool in 2007 called GOOG-411, which offered free directory assistance by voice. The service was discontinued in 2010 without an official explanation from Google, but some speculate that it was shut down because the company had already achieved its underlying goal: collecting a sufficient database of voice samples to advance its speech recognition technology.

At the time, Google VP Marissa Mayer said it outright: “The speech recognition experts that we have say: If you want us to build a really robust speech model, we need a lot of phonemes, which is a syllable as spoken by a particular voice with a particular intonation. So we need a lot of people talking, saying things so that we can ultimately train off of that. ... So 1-800-GOOG-411 is about that: Getting a bunch of different speech samples so that when you call up or we’re trying to get the voice out of video, we can do it with high accuracy.”

OpenAI spokesperson Taya Christianson said the company won’t be using these calls to train large language models.

OpenAI’s cofounder and former chief scientist, Ilya Sutskever, made headlines earlier this year after he left to start his own AI lab called Safe Superintelligence Inc. He has avoided the limelight since his departure but made a rare public appearance in Vancouver on Friday at the Conference on Neural Information Processing Systems (NeurIPS).

“Pre-training as we know it will unquestionably end,” Sutskever said onstage. This refers to the first phase of AI model development, when a large language model learns patterns from vast amounts of unlabeled data — typically text from the internet, books, and other sources.

During his NeurIPS talk, Sutskever said that, while he believes existing data can still take AI development farther, the industry is tapping out on new data to train on. This dynamic will, he said, eventually force a shift away from the way models are trained today. He compared the situation to fossil fuels: just as oil is a finite resource, the internet contains a finite amount of human-generated content.

“We’ve achieved peak data and there’ll be no more,” according to Sutskever. “We have to deal with the data that...

Read the full story at The Verge.

The lawsuit between Elon Musk and OpenAI is really heating up.

OpenAI just dropped a new blog post defending itself against Musk that outlines some new text messages between cofounders Ilya Sutskever, Greg Brockman, Sam Altman, Elon Musk, and former board member Shivon Zilis.

“You can’t sue your way to AGI,” the OpenAI blog post reads, referring to artificial general intelligence, which Altman has promised soon. “We have great respect for Elon’s accomplishments and gratitude for his early contributions to OpenAl, but he should be competing in the marketplace rather than the courtroom. It is critical for the U.S. to remain the global leader in Al. Our mission is to ensure AGI benefits all of humanity, and we have been and will remain a mission-driven organization. We hope Elon shares that goal, and will uphold the values of innovation and free market competition that have driven his own success.”

Some of the new messages revealed show Brockman telling Zilis in July 2017 about a meeting he had with Musk, who allegedly said that a non-profit was definitely the right structure early on but “may not be the right one now.” Later that month, Brockman wrote to Musk that path for OpenAI should be: “1. AI research non-profit (through end of 2017) 2. AI research + hardware for-profit (starting 2018) 3. Government project (when: ??).”

The blog also highlights Musk’s attempts to maneuver into the CEO position and gain majority control of the company (though it adds that on one call Musk said he “didn’t care about equity” but “just needed to accumulate $80B for a city on Mars”). Musk also proposed that OpenAI spin into Tesla, which has been previously revealed. When the negotiations fell apart because OpenAI’s cofounders rejected his proposal (Brockman and Sutskever admitted they had fears of a power struggle), Musk resigned from the company.

The blog said that after Musk resigned, he hosted a goodbye all-hands with the team where he encouraged them to “pursue the path we saw to raising billions per year” and that “he would pursue advanced Al research at Tesla, which was the only vehicle he believed could obtain this level of funding.”

Later, around the time Musk was working to acquire Twitter, he texted Altman that he was “disturbed” to see the company’s new $20 billion valuation. “De facto. I provided almost all the seed, A and most of B round funding,” he wrote, according to the disclosed texts. “This is a bait and switch.”

A few months after that interaction, Musk started an OpenAI competitor, xAI.

Some of the messages published by OpenAI were previously outlined in court filings that Musk made in his ongoing suit against OpenAI and its partner Microsoft. The lawsuit, filed by Musk in March, alleges that OpenAI had strayed from its original nonprofit mission to develop AI for the public good (he withdrew it in June 2024 without explanation, then refiled in August 2024).

Today’s update from OpenAI attempts to counter Musk’s narrative by offering evidence that he, not Altman, attempted to seize control in the company’s early days — a direct response to Musk’s recent lawsuit claims about Altman’s power consolidation.

OpenAI worried that ChatGPT would be a dud two years ago. Now, the stakes have never been higher.

Read the full story at The Verge.

I 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 Munch’s The Scream, Georges Seurat’s Sunday Afternoon, and Hokusai’s The Great Wave off Kanagawa. “There’s some interesting pieces of art here,” said Bibo Xu, Google DeepMind’s lead product manager for Project Astra. “Is there one in particular that you would want to talk about?”

Project Astra, Google’s 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?”

I was at Google’s sprawling Mountain View campus, seeing the latest projects from its AI lab DeepMind. One was Project Astra, a virtual assistant first demoed at Google I/O earlier this year. Currently contained in an app, it can process text, images, video, and audio in real time and respond to questions about them. It’s like a Siri or Alexa that’s slightly more natural to talk to, can see the world around you, and can “remember”...

Read the full story at The Verge.

OpenAI launched Sora, its text-to-video AI model, on Monday as part of its 12-day “ship-mas” product release series, as The Verge previously reported it would. It’s available today on Sora.com for ChatGPT subscribers in the US and “most other countries,” and a new model, Sora Turbo. This updated model adds features like generating video from text, animating images, and remixing videos.

With a ChatGPT Plus subscription, OpenAI says you can generate up to 50 priority videos (1,000 credits) at resolutions up to 720p with 5-second durations. The $200 per month ChatGPT Pro subscription that launched last week comes with “unlimited generations” and up to 500 priority videos while bumping the resolution to 1080p and the duration to 20 seconds. The more expensive plan also allows subscribers to download videos without a watermark and perform up to five generations simultaneously.

OpenAI first teased its text-to-video AI model, Sora, in February, and earlier today, Marques Brownlee, aka MKBHD, confirmed the launch with a preview based on his experiences testing Sora so far.

During the livestream, OpenAI showed off Sora’s new explore page with a feed of AI-generated videos created by other community members. The company highlighted a feature called “storyboards” that let you generate videos based on a sequence of prompts, as well as the ability to turn photos into videos. OpenAI also demonstrated a “remix” tool that lets you tweak Sora’s output with a text prompt, along with a way to “blend” two scenes together with AI.

OpenAI says videos generated with Sora will have visible watermarks and C2PA metadata to indicate they’re made with AI. Before uploading an image or video to Sora, OpenAI prompts you to check off an agreement that says what you’re uploading doesn’t contain people under 18, explicit or violent content, and copyrighted material. It says the “misuse of media uploads” could result in an account ban or suspension.

“We obviously have a big target on our back as OpenAI,” OpenAI vice president of research Aditya Ramesh said during the livestream. “We want to prevent illegal activity of Sora, but we also want to balance that with creative expression. We know that... will be an ongoing challenge, we might not get it perfect on day one. We’re starting a little conservative, and so if our moderation doesn’t quite get it right, just give us that feedback.”

If you don’t have a ChatGPT subscription, you’ll still be able to browse through the feed of AI-generated videos created by other people using Sora. While the model will become available in the US and many other countries today, OpenAI CEO Sam Altman said that it may “be a while” for a launch in “most of Europe and the UK.”

The release of Sora comes just a week after a group of artists, who claimed to be part of the company’s alpha testing program, leaked the product in protest of being used by OpenAI for what they claim was “unpaid R&D and PR.”

Friend just raised another $5.4 million to bring AI friends into the real world.

Read the full story at The Verge.

OpenAI is planning to launch new features, products, and demos for 12 days straight — starting with the full release of its o1 reasoning model.

OpenAI’s 12 days of “ship-mas” have officially begun, with the company set to reveal some new features, products, and demos during all 12 days starting December 5th, just a few days shy of the second anniversary of ChatGPT’s explosive launch in 2022. According to the company, ChatGPT now has over 300 million weekly users.

The event started with the release of its improved OpenAI o1 reasoning model out of preview, along with a new $200 ChatGPT Pro subscription that offers unlimited access to GPT-4o, Advanced Voice Mode, and an exclusive version of o1 that Sam Altman says “can think even harder for the hardest problems.”

Sources tell The Verge that OpenAI will also launch Sora, its text-to-video AI generator. There’s likely a lot more to come, and we’re keeping track of all the announcements from OpenAI.

You can follow along below as we continue to report on all the latest updates.

OpenAI is creating a new, more expensive tier for its flagship chatbot ChatGPT, and is bringing its “reasoning” model series out of preview with an update.

The company is releasing the full version of its o1 model (replacing o1-preview), which was initially released as a limited preview in September (code-named ‘Strawberry’). The new model will be available for ChatGPT Plus and Team users today, while Enterprise and Edu users will have access to it starting next week.

The company is also introducing ChatGPT Pro, a new $200 monthly subscription tier that includes unlimited access to OpenAI o1, GPT-4o, and Advanced Voice mode. It also includes a version of o1, exclusive to Pro users, that uses more compute to provide the best possible answer to the hardest problems (called o1 pro mode). The company will continue to offer a Plus tier for $20 a month that includes early access to new features, access to all the company’s models (except the more powerful o1 version), and more.

OpenAI

The Verge previously reported on the startup’s plans to kick off a “shipmas” period of new features, products, and demos for 12 days, with announcements that’ll include OpenAI’s long-awaited text-to-video AI tool Sora and a new model.

The company said that compared to o1-preview, users can expect a faster, more powerful, and more accurate model that is better at coding and math. It can also provide “reasoning” responses to images. And OpenAI promises it’s been trained to be more concise, which should result in faster response times than o1-preview.

OpenAI plans to add support for web browsing, file uploads, and more in ChatGPT — though there’s no timeline for these changes.

It also announced a ChatGPT Pro Grant Program that awards 10 grants of ChatGPT Pro to medical researchers at leading institutions, with plans for additional grants across various disciplines.

OpenAI first teased its text-to-video AI model, Sora, back in February and hasn’t provided any meaningful updates on when it will be released since then. Now, it looks like some artists leaked access to the model in protest of being used by the company for what they claim is “unpaid R&D and PR.”

On Tuesday, a group of Sora beta testers claimed to have leaked early access to Sora with a working interface for generating videos. In a post on Hugging Face, a public repository of AI models, they say that people were able to create lots of AI videos — all of which resemble OpenAI’s own Sora demos — before the company intervened to shut down access. (TechCrunch first reported on the alleged leak.)

From the group’s open letter:

DEAR CORPORATE...

Continue reading…

A cozy workspace full of character and characters — and a cat.

Continue reading…

In a stunning misstep, OpenAI engineers accidentally erased critical evidence gathered by The New York Times and other major newspapers in their lawsuit over AI training data, according to a court filing Wednesday.

The newspapers’ legal teams had spent over 150 hours searching through OpenAI’s AI training data to find instances where their news articles were included, the filing claims. But it doesn’t explain how this mistake occurred or what precisely the data included. While the filing says OpenAI admitted to the error and tried to recover the data, what it managed to salvage was incomplete and unreliable — so what was recovered cannot help properly trace how the news organizations’ articles were used in building OpenAI’s AI models....

Continue reading…