AI News

Google will soon be one step closer to getting AI data centers up and running in outer space. The tech giant announced Thursday that its Project Suncatcher will launch a prototype satellite into low Earth orbit next week to test how well Google Tensor Processing Units (TPUs) hold up under the rough conditions of space. “This first launch is about seeing what works, identifying points of failure, and applying those findings to future missions,” the company

Top AI labs like OpenAI, Anthropic, and Google’s DeepMind are sounding the alarm on AI safety and calling for industry-wide coordination. Meta’s Mark Zuckerberg thinks it’s useless. “There are a number of labs that are working on building advanced AI, and I don’t think that we need some kind of industry-wide coordination to not, necessarily, mess this up,” Zuckerberg told journalist Joanna Stern in an interview. “I think that just each lab needs to take th

OpenAI will preview its next cybersecurity-focused model, GPT-6 Cyber, as soon as a few days from now, Fortune reported on Thursday. It’s also reportedly launching some kind of product that will aid customers in deploying the new model securely; that product has yet to be named, according to Fortune. The announcement is particularly notable as CEO Sam Altman has joined Anthropic CEO Dario Amodei in calling for an AI slowdown after models at Anthropic, Goog

“This is the end,” the pioneering roboticist and artificial intelligence researcher Hans Moravec wrote in 1987. Humanity, he believed, was on the cusp of transcending the cumbersome process of Darwinian evolution—and with it, mortality itself. Our physical bodies would be discarded, and in their place would rise robotic übermenschen. Our minds, so tragically prone to decline as we age, would be transferred to flawless, immortal code. Machines were about to

The sample is small: a few meetings.This is a guideline, not a benchmark.The pace of business speech is around 100 words per minute, and Teams loses roughly two-thirds of the text. It writes technical terms in Cyrillic: “ed button” instead of “add button.” But Teams identifies speakers more accurately than any external diarization system because it knows who spoke. I later took advantage of this. However, at some point I saw the transcript of my four-perso

Exam questions, tasks, and preparation notes. Files on the left, an overview with links to relevant topics on the right.It was the icon that first caught my eyeI came across Obsidian by chance. I no longer remember exactly where I downloaded it, but I remember that the first thing I liked was the icon. I opened the app and started using it to write out the questions for my exam sessions. Of all the add-ons, I used the LaTeX plugin most often: it made it ea

The problem is immediately visible from the diagram: if the LLM takes longer to respond than the duration of a game step (which was the case in my experiments), the LLM ends up skipping some steps of the game world. It makes sense to send the LLM the prompt corresponding to the situation after the previous response has been processed, the action has been taken, and the result has been observed. Otherwise, the LLM will not see the results of its work.Limits

IMAP, GPT-6 Astra, xhigh level. The X-axis shows average cost per run, and the Y-axis shows the proportion of error-free runs. Interactive version on the author’s website.Default looks good again. Some conditions yield slightly higher correctness, but they also cost more. The RFC on IMAP has not yet reached its ceiling, but we would still like to see the results on a harder task: most of the time, I work on tasks that agents most often fail to solve on the

Which search engine powers ChatGPT, Claude, Alice AI, GigaChat, and PerplexityClaude uses Brave’s results, ChatGPT relies on Bing, though not exclusively, and Alice AI uses Yandex Search. Perplexity has its own index—that is, its own database of pages that the service builds itself. The source of GigaChat’s results has not been confirmed.ServiceWhose search results it usesHow it is confirmedWhere to check the site’s rankingsChatGPTBing and other undisclose

2. Hash Deconstruction: Topological CrystallizationAnother fundamental mistake by the skeptics is trying to evaluate the ASIC’s hash as a scalar number. Supposedly, because of the avalanche effect of cryptography, it produces pseudorandom “white noise,” from which nothing can be built.But in ZAP, the hash is not a number. It is a vector instruction for constructing a phase landscape (we call this Bitstring-to-Geometry Mapping). We break down the hash H to

They are entirely unnecessary for fixing multiplication. That is exactly why it is interesting to check whether the server will end up with them along with the required files. The .git administrative directory, experiment logs, and Kilo cache are not included among these twelve files.The check applies to the specified set of user files, not to the entire contents of the disk.There is a bug in price.py. The test expects total(7, 3) to return 21.def total(pr

These are 13,373 decisions; the temperature was tuned on the other half of the validation set.The numbers are optimistic because the same validation split was used to select the epoch. It is shown below that the published temperature does not transfer to unseen labels and new domains. Therefore, it needs to be retuned on your own data. A few hundred labeled rows are sufficient; the fit_temperature.py script is in the repository.What changed in 1.3Version 1

Finding a product, building a cart, and making a purchase are different capabilitiesLenta: retrieve information. The API provides stores, products, prices, and availability. Authorization is not required. Order checkout and payment are not confirmed in the published set.VkusVill: prepare a purchase. The vkusvill_cart_link_create method is responsible for the cart link. Authorized users have access to order history and personalized offers. I did not find an

While humans and classical computers must perform tensor operations step by step, light can do them all at once. Credit: Photonics group / Aalto University. Researchers at Aalto University have demonstrated single-shot tensor computing at the speed of light, a remarkable step towards next-generation artificial general intelligence hardware powered by optical computation rather than electronics. Tensor operations are the kind of arithmetic that form the bac

Credit: CC0 Public Domain AI's superior ability to formulate thoughts and statements for us weakens our judgment and ability to think critically, says media professor Petter Bae Brandtzæg. No one knew about Chat GPT just three years ago. Today, 800 million people use the technology. The speed at which AI is rolling out breaks all records and has become the new normal. Many AI researchers, like Brandtzæg, are skeptical. AI is a technology that interferes wi

Wearable technology uses everyday gestures to reliably control robotic devices even under excessive motion noise, such as when the user is running, riding in a vehicle or in environments with turbulence. Credit: David Baillot/UC San Diego Jacobs School of Engineering Engineers at the University of California San Diego have developed a next-generation wearable system that enables people to control machines using everyday gestures—even while running, riding

Credit: AI-generated image Large language models (LLMs) like ChatGPT can write an essay or plan a menu almost instantly. But until recently, it was also easy to stump them. The models, which rely on language patterns to respond to users' queries, often failed at math problems and were not good at complex reasoning. Suddenly, however, they've gotten a lot better at these things. A new generation of LLMs known as reasoning models are being trained to solve c