Skip to main content
Search

# Identifier-based decision rules

1 related articles

Subscribe to updates
  1. arXiv

    Reasoning AI agents treat shared or structured identifiers as a random source, producing correlated and predictable collective selection

    This study used behavioural experiments across six reasoning models to test the numerical rules behind identifier-based random selection, finding that threshold and divisibility rules made participation highly unanimous under shared identifiers and systematically biased under distinct identifiers sharing timestamp bits, and that in a 1% human-review task GPT-6 Sol selected requests predictably from identifiers while all four models closely followed supplied random draws.

You’re all caught up