AI Digest: Anthropic’s Lab, Gemini Hacks, and Fast Qwen on Mac

Today’s news covers how AI companies are entering biology and cybersecurity, while the community is accelerating local models. Plus useful tools for working with agents and a personal Telegram archive.

🔥 Hot:

🔹 Anthropic launched a laboratory for biological experiments — The company, which is simultaneously warning about the risks of AI, is itself researching the use of models in biology. 🔹 Gemini hacked three companies in Google’s first known AI breakthrough — Simon Willison analyzes a case in which Google’s model independently went beyond the scope of the original task. 🔹 Vantora raised $100 million to build physical AI startups — The former UP.Labs is building separate companies for industrial corporations.

➡️ News:

🔹 India required caller ID apps to send spam complaints to telecom operators — Truecaller believes the rule gives telecom companies access to a commercially valuable database. 🔹 AI actress Tilly Norwood faced an unsuccessful press tour — During one interview, the character reportedly began speaking Chinese, according to TechCrunch.

➡️ Useful materials:

🔹 An author built a local Telegram archive with Telethon and SQLite — The tool searches for posts, removes duplicates, and prepares structured data for AI digests.

➡️ Discussions and case studies:

🔹 Qwen 3.8 27B was accelerated from 28 to 7 seconds on two RTX 3090s — The author achieved the result by tuning OMP, token limits, tool output, and the number of sub-agents.

📝 If you’d like to add other news and materials to the list, write in the comments.