• Home
  • News
  • Blog
  • Releases
  • LLM history
  • Compare LLMs
  • Library
  • About
⌘K
Sign in

A blog and notes on development. The easiest way to reach me is via the social links below.

Contacts
talalaev.misha@gmail.com
Documents
Personal data processing policyPersonal data processing consent
Photo: Luiz Felipe / Unsplash

Anthropic sent the police a false tip, while Deno awaits a development halt.

Sh0ny
Sh0ny
10 October 2026
  1. Home
  2. Blog
  3. Anthropic sent the police a false tip, while Deno awaits a development halt.
2 min read

In short

Today’s roundup: AI system errors with real-world consequences, developer news, and research on the limits of AI agents. Plus, mathematicians are trying to make sense of the stream of results from OpenAI.

Today’s roundup: AI system errors with real-world consequences, developer news, and research on the limits of AI agents. Plus, mathematicians are trying to make sense of OpenAI’s flood of results. 🔥 Hot: 🔹 An Anthropic model sent the Philadelphia police a false homicide tip: The message went to spam, and the company only learned about it some time later. 🔹 Mathematicians are trying to make sense of OpenAI’s massive collection of results: Researchers call the publication’s volume staggering, while what it all means remains unclear. 🔹 Cloudflare acquired the company developing Deno and plans to stop developing the platform: They promise to release security updates for another year, after which the community may take over its development. ➡️ News: 🔹 The creator of the Jev AI model was valued at $7.5 billion just weeks after launch: TypeSafe claims that Jev works faster and uses fewer tokens than LLMs. 🔹 Anthropic disconnected its internal AI agent tests from the live internet: The company said it will not restore access for now. ➡️ Useful resources: 🔹 Study: AI agents write more code but do not create more software: Reviewing and fixing generated code is still a hassle. 🔹 AMBER teaches web agents to remember what matters in long tasks: The authors propose append-only memory in which the agent stores facts, errors, and corrections. 🔹 Google DeepMind and Biohub explain why AlphaFold did not solve the protein-folding problem: The conversation discusses why scaling computation and data alone may not be enough. 🔹 An analysis of Claude’s nine-loop calculation shows how to verify scientific code: The author describes a process decomposition but discovers an error in their own calculations. ➡️ Discussions and case studies: 🔹 Nikon stripped the creator of a generative AI-edited video of their victory: The company decided that the video did not comply with the contest rules. 🔹 A Habr author shows how to communicate with AI via SMS: This method can be useful where mobile internet is unstable. 📝 If you would like to add other news and resources to the list, write in the comments.

Newsnewsai
More AI-tool write-ups on the Telegram channel — short and to the point
Subscribe

Comments

(0)
​