• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI
    AI
    Report

    Astra reportedly almost always detects tampering with its trajectory in updated CIAware-Bench tests

    The poster argues that retained reasoning from past turns makes omissions at intervention steps incriminating for DtT or TE protocols.

    Maksym AndriushchenkoMA
    Alexander PanfilovAP
    2 Sources, ,

    TLDR

    A poster says their group updated CIAware-Bench with recent frontier models and found Astra almost always figures out whether its trajectory was tampered with. They argue that, when reasoning from past turns is retained, its omission at an intervention step makes DtT or TE protocols incriminating. They recommend designing protocols around an assumption of perfect intervention awareness.

    Combined views

    724

    2 Sources, first seen 2h ago

    Combined views

    724

    2 Sources, first seen 2h ago

    19 likes
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet

    2h ago
    first seen 2h ago
    19 likes
    1 comments
    5 saves
    3 reposts
    1 comments
    5 saves
    3 reposts

    2 Sources

    Alexander Panfilov@kotekjedi_mlWe have updated CIAware-Bench with recent frontier models and found that Astra basically saturates it and almost always figures out whether its trajectory was tampered with. In realistic deployments, where we do not omit reasoning from past turns, DtT or TE protocols become very incriminating because reasoning is omitted at the intervention step. We argue that we should assume perfect intervention awareness when designing protocols from now on.2h
    Maksym Andriushchenko@maksym_andrRT @kotekjedi_ml: We have updated CIAware-Bench with recent frontier models and found that Astra basically saturates it and almost always f…2h
    Featured Source

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    2 Sources

    Alexander Panfilov@kotekjedi_mlWe have updated CIAware-Bench with recent frontier models and found that Astra basically saturates it and almost always figures out whether its trajectory was tampered with. In realistic deployments, where we do not omit reasoning from past turns, DtT or TE protocols become very incriminating because reasoning is omitted at the intervention step. We argue that we should assume perfect intervention awareness when designing protocols from now on.2h
    Maksym Andriushchenko@maksym_andrRT @kotekjedi_ml: We have updated CIAware-Bench with recent frontier models and found that Astra basically saturates it and almost always f…2h