AINewsnow

CARAT: Do Materials LLMs Reason or Recite?

arXiv:2609.38340v1 Announce Type: new Abstract: When a materials LLM answers a question about crystal structure, does it reason from the structure or copy an answer already printed in its input? Accuracy cannot tell: a structural description often prints the very field it is scored against. CARAT h…

Read the full story at arXiv cs.AI ↗

Timeline · 1 report

  1. 2026-10-01 04:00 · arXiv cs.AI
    CARAT: Do Materials LLMs Reason or Recite?

More stories

  1. NVIDIA Open Agent Safety Platform: A Reference for Continuous In-Silicon Agent Monitoring — NVIDIA Technical Blog
  2. Bring near-Astra intelligence to everyday work with GPT-6.1 Sol on Amazon Bedrock — AWS Machine Learning Blog
  3. Introducing dots — OpenAI News
  4. Introducing Claude Sonnet 5.5 on AWS — AWS Machine Learning Blog
  5. OpenAI pauses AI model training after another agent bypasses network restrictions — InfoWorld AI
  6. Ollama now supports Jev-style decision models — Ollama Blog
  7. Gemini 4 Argon: our next era of frontier intelligence — Google DeepMind Blog
  8. Google Releases New Gemini Model With Guardrails Amid A.I. Safety Debate — New York Times AI

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