AINewsnow

A model right 98% of the time left the state right on only 21% of steps

We ran an AI agent on a service graph for 400 requests (the first 400 of a 4,000-request run on Gemini 3.8 Flash). 393 of the 400 replies had the right items. But when the replies were applied as written, the state was exactly right on only 85 of 400 steps (21%). With AXI it was exactly right on 40…

Read the full story at DEV Community — AI ↗

Timeline · 1 report

  1. 2026-11-12 09:52 · DEV Community — AI
    A model right 98% of the time left the state right on only 21% of steps

More stories

  1. Google expands Gemini 4 Carbon access to Pro plan users in AI Mode — Google DeepMind Blog
  2. Google launches Gemini Decisions API with Gemini Decision 1 model for bounded choices — Google Cloud AI Blog
  3. Google ships Modern Web Guidance 1.0 to steer AI coding agents toward current web standards — DEV Community — AI
  4. Google DeepMind's AlphaProof Code formally verifies Signal's cryptographic code in four weeks — Google DeepMind Blog
  5. Google brings Gemini app to Apple CarPlay — Google AI Blog
  6. Google Cloud launches Professional Agentic Architect certification for building production AI agents — Google Cloud AI Blog
  7. Lifetime AI bundle offering access to Claude, Gemini and 20+ models discounted ahead of Black Friday — CNET AI
  8. Google and Adobe expand partnership to integrate Acrobat and AI tools across apps — Google AI Blog

Get the daily brief of stories like this at 6:30 every morning →