Karpathy Endorses 'Profanity Index' for Model Eval

Andrej Karpathy validates the newly released 'profanity index', a token tracker capturing user frustration, considering it a potentially strong evaluation signal for reflecting intuitive emotional feedback in AI interactions.

2026-07-09 ~ 2026-07-09 · 2 related posts

• Karpathy on the 'Swear Word Index' as an Eval Signal — karpathy · 2026-07-09
• Dev Launches Token Tracker 'Swear Word Index' — ptkbhv · 2026-07-09