shipfeedAI news, curated daily

10:45:21 CET
26 JUL10:45:21shipfeed⋯
pull to refreshlast sync
Just in — 30 new
§ safety · storyline

Adversarial pragmatics benchmark tests AI safety against instruction

Adversarial pragmatics benchmark tests AI safety against instruction

Jul 1 · · primary fetch1 sourceupdated Jul 1 ·

This storyline groups 1 article from 1 source. The originating feed didn’t ship an excerpt — open any link below to read the piece.

read full article on arxiv.org ↗
§ sources1 publication · timeline below
  1. arxiv.orgAdversarial Pragmatics for AI Safety Evaluation: A Benchmark for Instruction Conflict, Embedded Commands, and Policy Ambiguityprimary