Archived edition

AI news

5 stories that day · 1 model · 2 open source · 1 research · 1 trending

Published daily at 7 PM ET · Updated · RSS

Also that day

  1. Model

    Gemini can now create, preview and edit documents, PDFs and slides right in the chat (opens the original source in a new tab)

    You can type changes yourself, highlight text to ask about it, or have Gemini rework the file. Google says PDFs work on web and phone, and Docs, Sheets, slides and code on the web only.

    Google · Original source (opens in a new tab)

  2. Trending

    A developer finds Claude Code re-sends background agents' full transcripts just to fill a status line (opens the original source in a new tab)

    Flamerix AI says that about every 30 seconds Claude Code sends each background agent's whole transcript to the model to write its short status, and shares a script that stops this without changing the app.

    562 points on Hacker News

    Flamerix AI · Original source (opens in a new tab)

  3. Research

    The Trinity lets AI agents improve their own tooling by practising in simple made-up worlds (opens the original source in a new tab)

    Three roles group tasks by the skills they need, build small practice worlds for each skill and evolve reusable agent modules. The authors report average gains of 34 points on training tests and 26 on new ones.

    386 upvotes

    Paper · Original source (opens in a new tab)

  4. Open source

    PageIndex 0.5.0 lets AI answer questions from long documents without a vector database (opens the original source in a new tab)

    The release adds a faster default engine that builds each document's tree from its layout, and a local mode that needs no server, vector database or PageIndex API key. It also runs on a Claude subscription.

    VectifyAI · Original source (opens in a new tab)

Recent highlights by topic

Notable stories from the past few days. Each shows its own date.

Models and launches

Frameworks and open source

Research to watch

Trending on GitHub