OpenAI's new GPT-6.2 Astra: by OpenAI's own tests, it beats its previous Astra model at long, multi-step office work for under a third of the cost. The real story may be price. Europe note: businesses can keep API data in EU regions from day one. https://t.co/1tvccPF8hO
A physicist says OpenAI's GPT-6 Astra led the detailed proof behind a candidate way to define the whole Standard Model of particle physics on a grid, filling a gap open for 25+ years. He's a visiting scientist at OpenAI, and experts still need to check it. https://t.co/MQaaLG8fIJ
OpenAI's AI agents can now pay for things for you. You approve each purchase with Face ID or a fingerprint, and the shop never sees your real card details. For now it's US only. In Europe, the agents themselves aren't yet offered on personal Pro plans. https://t.co/FnXohQOXUz
EU leaders held their first AI summit on Friday in Dublin, with Mistral's Arthur Mensch on competitiveness and Geoffrey Hinton on the risks. Investment, energy and less reliance on others were the themes. The test is whether money and decisions follow. https://t.co/5PBsRHrKOj
Anthropic says about 1,000 copies of its new Claude Fable 5.2 AI turned the proof of Fermat's Last Theorem into computer-checked form in 30 hours. An earlier Claude attempt, guided by a human, took 11 days. Company claim, not yet independently reviewed. https://t.co/31qFmXP1uh
@MistralAI Nice real-world test. How much did feeding in speaker names and technical terms cut the error rate versus the plain model? Names and jargon are usually where live captions fall apart.
Visa had 22 AI models act as shopping assistants: talk to a customer, work out what they need, then buy or walk away. They passed 57% of runs, and in about half of the failures the AI still bought something. https://t.co/XR93ZfZOxy
DeepMind: two AI agents redid a known 1024-bit RSA computation with 3.5x less computing time, mainly via better code. This result doesn’t show a break of modern 2048-bit RSA. Working paper, not yet peer reviewed. https://t.co/zw9pZkwssc
MIT CSAIL: GPT-6 Astra edited 0.016% of a small Qwen model's weights, no training, and its math score went from 7.9% to 48.1%. Catch: it fixed reasoning loops, not skill. An independent re-run gets ~44% with Qwen's own settings and no edit. https://t.co/YVUonoactx
First generative-AI case at the EU Court of Justice: the Advocate General's non-binding opinion says training on press articles is generally covered by the data-mining exception, but chatbot answers that replace the article may infringe publishers' rights. https://t.co/4vK3lMOkCK
Anthropic is extending its invisible text watermark to older Claude models, under its EU AI Act commitments, and applies it worldwide. Detection is limited for now to regulators, researchers and similar groups. Curious how well it survives real editing. https://t.co/kKB5YeE5Lj
@AnthropicAI Useful complement to the text watermarks you're adding for the EU AI Act: those mark that text came from Claude, while this proves which model produced a reply. At ~70-95x the GPU time today, is the aim audits, or proofs on every API call one day? https://t.co/ciZoXxYTEI
Nice to see from Europe: on 29 Oct Mistral released the open weights of Mistral Large 4, a 1-trillion-parameter model it says was trained from scratch on European infrastructure. Curious to see how teams put it to work. https://t.co/wBQEGqFozN
OpenAI and Anthropic each claim a Lean-checked AI proof that the Exponential Time Hypothesis is false. Neither is peer reviewed or fully checked by a human yet. An open review of OpenAI's, led by Impagliazzo, Paturi and Williams, is underway. https://t.co/4XWouyp76j
We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics.
The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra.
The problem concerns whether the description of smooth three-dimensional fluid motion modeled by the Navier-Stokes equations can break down. It has remained unresolved for roughly 90 years.
Remember that robot with very realistic face expressions? Now it talks too. No pre-script. It's :o
https://t.co/68KQ8JOzaU
#tech#humanoids#robot#ameca#gpt3#future#TechNews
Video direct link: https://t.co/4ib5kocu1v
Article: