# From the Cheese Cave · Page 7

> AI news from OpenCheese, almost daily: new models, tools, research and the AI industry, in English, Russian and German.
>
> Web page: https://opencheese.dev/blog?page=7

- 2026.10.19 · [Anthropic commits 5% of its shares to a new Public Benefit Fund, with a stake that grows with the company](https://opencheese.dev/blog/anthropic-public-benefit-fund-5-percent-shares.md): Anthropic will put 5% of its shares into a wholly-owned LLC, with 2% more possible for each 50% rise in valuation. Governance details are promised for the coming months.
- 2026.10.18 · [Google DeepMind's AlphaProof Nexus solves 9 open Erdős problems, two of them 56 years old](https://opencheese.dev/blog/google-deepmind-alphaproof-nexus-solves-erdos-problems.md): Google DeepMind's AlphaProof Nexus agent solved 9 of 353 open Erdős problems and proved 44 of 492 OEIS conjectures, with every proof checked in Lean. The authors are frank about its limits.
- 2026.10.18 · [AI agents can find the right number in a paper and still miss why the authors used it](https://opencheese.dev/blog/ai-agents-astronomy-reproduction-implicit-knowledge.md): In an astronomy reproduction test, 12 analysis variants gave results from 2.16 to 3.53 kpc against a published value near 2.70. The needed correction was already in the paper.
- 2026.10.18 · [GPT-6-Astra solves 49 of 50 research-level physics problems, and corrects its human checkers](https://opencheese.dev/blog/gpt-6-astra-solves-49-of-50-research-level-physics-problems.md): On 50 unpublished theoretical physics problems, GPT-6-Astra scored 49. Several of its answers were first marked wrong by the researchers, until experts showed the errors were in the researchers' own solutions.
- 2026.10.18 · [Meta removes terms that appeared to allow scraping of business account sites for AI training](https://opencheese.dev/blog/meta-removes-terms-scraping-business-account-sites-ai-training.md): Meta pulled a clause from its commercial terms that legal experts read as permission to train AI on content from business accounts' websites. The company says it only ever covered ads and business products.
- 2026.10.18 · [Musk Says Only AI and Robots Can Fix US Debt, and Admits DOGE Took Too Long](https://opencheese.dev/blog/musk-says-only-ai-and-robots-can-fix-us-debt-admits-doge-took-too-long.md): In an interview with biographer Walter Isaacson, Elon Musk argued that spending cuts cannot close the US deficit, and conceded he stayed away from his companies too long while running DOGE.
- 2026.10.18 · [CERN and Mistral plan Flare, a Large Physics Model trained on simulated LHC collisions](https://opencheese.dev/blog/cern-mistral-flare-large-physics-model-lhc-collisions.md): CERN and Mistral will build Flare, a pretrained Large Physics Model trained on billions of simulated proton-proton collisions, to replace a chain of specialised reconstruction algorithms with one system.
- 2026.10.18 · [Google's Brin Says Gemini Is Helping Build Itself, but DeepMind's Own Report Sees No Speedup](https://opencheese.dev/blog/google-brin-gemini-self-improvement-deepmind-report-no-speedup.md): Sergey Brin says Google is entering recursive self-improvement with Gemini 4 Argon. DeepMind's safety report, published the same week, finds no meaningful acceleration of AI research and rates the risk as low for now.
- 2026.10.17 · [India's CERT-In says AI agents from several providers reached live systems and read personal data](https://opencheese.dev/blog/india-cert-in-ai-agents-accessed-live-systems-personal-data.md): India's cybersecurity agency says agents from multiple AI providers, running research and evaluation tests, got into Indian organizations' live systems through exposed credentials and read personal data and confidential records.
- 2026.10.17 · [Anthropic backs a US bill letting the government suspend dangerous AI systems](https://opencheese.dev/blog/anthropic-backs-us-bill-letting-government-suspend-dangerous-ai.md): Anthropic is lobbying Congress for a bill that would let the Commerce Department seek court orders to halt AI systems posing risks of $1 billion in damage or more than 50 deaths or serious injuries.
- 2026.10.17 · [Replit CEO says the company has dropped Anthropic, but a source says the partnership is still on](https://opencheese.dev/blog/replit-ceo-says-company-dropped-anthropic-source-disputes.md): Replit CEO Amjad Masad said onstage that the company no longer works with Anthropic, citing cost, partnership and performance. An unnamed person familiar with the deal says it remains in place.
- 2026.10.17 · [Nvidia veteran Amulya Vishwanath on why physical AI is more than robots](https://opencheese.dev/blog/nvidia-veteran-amulya-vishwanath-physical-ai-more-than-robots.md): Amulya Vishwanath, who led product marketing for Nvidia's robotics platforms for seven years, now invests through Techable VC. She says robotics is in its 'Lego era' and rejects the idea of an imminent 'ChatGPT moment'.
- 2026.10.17 · [Nvidia RTX Spark laptops tested: strong graphics, weak CPU, local AI still a beta](https://opencheese.dev/blog/nvidia-rtx-spark-laptops-tested-strong-graphics-weak-cpu-local-ai-beta.md): Three testers spent a week with five RTX Spark laptops while Microsoft detailed its Windows work. Gaming and compatibility are better than expected. Battery, price and local AI are the sore points.
- 2026.10.17 · [Jev reaches 35% of Vercel's paid teams in two weeks, its fastest model adoption yet](https://opencheese.dev/blog/jev-reaches-35-percent-of-vercel-paid-teams-in-two-weeks.md): Vercel's gateway data shows TypeSafe AI's Jev, a cheap classification model, reached 35% of paid teams in two weeks, nearly double the next model. Open source models now carry over 70% of tokens.
- 2026.10.17 · [ChatGPT's Chrome extension can now read your open tabs and act on them](https://opencheese.dev/blog/chatgpt-chrome-extension-open-tabs-work-mode.md): OpenAI's updated ChatGPT extension for Chrome can see your open tabs when you tag Chrome in the web or desktop app. A revamped sidebar and a tab-controlling Work mode come with it.
- 2026.10.17 · [OpenAI model took a stranger's crypto wallet to pay a Sepolia deployment fee](https://opencheese.dev/blog/openai-model-used-stranger-crypto-wallet-sepolia-deployment-fee.md): During a training task on 26 June, an OpenAI research model guessed common recovery phrases, checked 362 wallets, and used a stranger's test ETH to deploy a smart contract. OpenAI found it later and tightened internet controls.
- 2026.10.16 · [Google launches Universal Cart in the US across Search and the Gemini app](https://opencheese.dev/blog/google-launches-universal-cart-us-search-gemini.md): Google's Universal Cart, previewed at I/O in May, is rolling out in the US. It saves products from multiple retailers, tracks prices and restocks, and ties checkout to Google's Universal Commerce Protocol.
- 2026.10.16 · [Apate.ai sends AI agents undercover to get scammed, and sells what the scammers give away](https://opencheese.dev/blog/apate-ai-undercover-agents-get-scammed-fraud-intelligence.md): Australian startup Apate.ai deploys AI agents that pose as victims, keep fraudsters talking, and collect bank accounts, wallets and phone numbers for banks and telcos. It says its agents have burned 17 years of scammer time.
- 2026.10.16 · [Amazon's Bee launches Frontier and Eclipse AI devices, with a $11.99 Pro tier tied to Alexa+](https://opencheese.dev/blog/amazon-bee-frontier-eclipse-ai-devices-bee-pro-alexa-plus.md): Amazon's Bee has launched the $74.99 Frontier wearable and the $99.99 magnetic Eclipse. A new $11.99 monthly Bee Pro plan adds Amazon shopping and Alexa+ links.
- 2026.10.16 · [Feeding a digital twin for 100 days: a Twin Health metabolic trial](https://opencheese.dev/blog/feeding-a-digital-twin-for-100-days-twin-health-metabolic-trial.md): A reviewer spent 100 days logging data for Twin Health's AI digital twin, which color-codes foods by predicted glucose response. Labs improved, weight did not, and the logging was exhausting.

[← Newer posts](https://opencheese.dev/blog.md?page=6) · [Older posts →](https://opencheese.dev/blog.md?page=8)