#нейросети
-
Mistral releases the weights of its flagship Mistral Large 4, aka "Le Chonk"
Mistral AI has published the weights of Mistral Large 4. It is a one-trillion-parameter multimodal model that the company calls the best open-weight model for coding and defensive cybersecurity.
-
A computer in the weights: a working RISC-V CPU compiled into Qwen 3.5
Developer Divyansh Agrawal has compiled a 32-bit RISC-V processor directly into the weights of Qwen 3.5 4B. The model's own Gated DeltaNet layers serve as RAM, and no external tools are involved.
-
A decade-old CNN beats frontier LLMs at diagnosing skin cancer
A British dermatologist pitted GPT-6 Astra and Claude Fable 5.1 against an old Inception v3 network fine-tuned in twenty minutes. On 200 biopsy-proven skin lesions, the old network was markedly more accurate.
-
Meta's new terms may let it train AI on news sites for free
Meta's revised commercial terms, taking effect on October 30th, appear to let the company crawl websites "associated" with business accounts, including news publishers, and use their content to train its AI.
-
AI Models Can Provably Hide Arbitrary Capabilities
Researchers show how an encrypted circuit can be buried inside a network's weights, firing only on a secret trigger — and staying invisible to every standard audit.
-
Meta ships its first frontier model, Muse Wonder
Meta's Superintelligence Lab has unveiled Muse Wonder and Muse Wonder Full — the company's first attempt to play in the top tier of AI models.
-
Anthropic Releases Sonnet 5.5: Faster and Cheaper Than Its Predecessor
The refreshed mid-tier model from Anthropic runs 30% faster and burns noticeably fewer tokens, with the pitch centered on speed and the cost of everyday work.
-
The Neural Network That Decides Whether a Number Is Numberwang
Developer GraafHenk has posted a tiny neural network on GitHub that delivers a verdict for Numberwang, a fictional game from a British sketch show. The whole project is a 1.8 MB model file and about a hundred lines of pure Python.
-
Cracking Open the Apple Neural Engine: Why the GPU Beats the NPU on Transformers
Developer Eileen Yoon returned to a project she had abandoned three years ago and finally reverse-engineered the full architecture of Apple's neural module on the M1 chip — just as Apple appears to be burying the standalone NPU.
-
DeepSeek Readies V4.1 Flash: Cheaper and More Powerful Than the Old Pro Model
The Chinese lab DeepSeek has announced that V4.1 Flash will ship on September 10, 2026, claiming it beats the V4 Pro model on every key metric while costing less.
-
US Intelligence Agencies Accuse Chinese AI Firms of Industrial-Scale “Distillation” of Western Models
The NSA, CISA and FBI have issued a joint bulletin claiming that Chinese developers are systematically siphoning off the capabilities of Claude, GPT, Gemini and Grok in order to train their own models more cheaply and quickly.
-
Why AI-Generated Food Looks Inedible
Restaurants and brands increasingly advertise their dishes with images from neural networks — and end up with wormlike noodles, ice cream made of concrete, and burgers that look assembled from rocks. The Verge set out to explain why this happens.
-
Z.ai Releases GLM-5.3 Weights — With License Caveats
The Chinese lab Z.ai has open-sourced the weights of GLM-5.3, which it calls its most powerful model for agentic coding and cyberdefense. This time, though, there won't be a fully free license.
-
Google Releases Gemini Omni 1.1 Flash: More Control Over Video Generation
Google has unveiled an update to its generative video model, Gemini Omni 1.1 Flash. Developers get scene extension, keyframe control and 4K output.
-
Qwen prepares the Flash-Next model: 125 billion parameters with 6 billion active
A page for a new language model in the Qwen family — Flash-Next — has been spotted on the ModelScope platform; judging by the name, its release is expected very soon.
-
GPT-5.6 Sol: OpenAI Finally Steps Up Its Computer Vision
The Roboflow team ran OpenAI's new lineup of models through its own benchmark for visual tasks. The Sol model turned out to be the company's best ever at “vision,” though it still falls short of the market leaders.
-
AI Isn't Outthinking Mathematicians — It's Out-Remembering Them
Davide Piffer offers a less flashy explanation for AI's successes in mathematics: the point isn't superior intellect, but an almost limitless symbolic working memory.
-
Z.ai Announces GLM-5.3 with a Focus on Coding and “Cyber Skills”
Z.ai has unveiled its GLM-5.3 language model, positioning it as a frontier system for writing code and crediting it with newly emerged “cyber skills.”
-
Google DeepMind unveils Gemini 3.7 Flash — half the price of its predecessor
Google DeepMind has released Gemini 3.7 Flash, an update to its coding-and-agents “workhorse” that arrived just three weeks after version 3.6 and costs half as much.
-
DeepSeek V4 Pro 0813: A Release with No Announcement and Mysterious Pelicans
The Chinese lab DeepSeek has released a new model, V4 Pro — so far only via the API and without an official announcement page. Simon Willison drew attention to several curious details of the launch.
- More