Tag
This paper describes two pre-specified empirical tests to evaluate whether phase-derived features improve prediction accuracy for commitment or contradiction endpoints, but results show no significant improvement over baseline models.
Microsoft's developer advocate tests whether agents work better with JSON payloads or conventional CLI arguments, finding that flat CLI args actually perform better in terms of success rate, token efficiency, and latency.
A detailed technical critique finds that Meta's Stable Signature, along with Google's SynthID and Adobe's TrustMark, have far lower detection accuracy than claimed, with high false-positive rates and unreliable watermarks.