Young Tries AI

Everything on Anthropic

3 mentions, newest first.

  1. Claude Mythos access reportedly opens to individual researchers via KYC

    A Reddit user reports that Claude Mythos access is now open to individual researchers through KYC. Enrolling used to require CVE or business verification. The listed steps: have a Claude subscription or API account, pick Cyber Verification Program in the Anthropic portal, choose individual, and apply for defense access.

    Partly usable
  2. Anthropic's engineering post defines agent memory and dreaming for Claude

    Anthropic defines agent memory as knowledge Claude gains from experience and retains between sessions, saved as context it can read later. It excludes weight changes, in-context learning and compaction. Dreaming means reflecting on logs between sessions to fix mistakes or produce new notes and representations.

    Not yet
  3. Anthropic publishes FlashProve research on proving which model produced an output

    Anthropic published research on FlashProve, an open source system for provable inference: a cryptographic proof that an output came from a specific model. It requires labs to publish a fingerprint of each model, which does not reveal weights. A technical specification and working code are on GitHub.

    Not yet