← Index

VERIFYDevtools & code

Jev Model Vetting

clay shentrup 🇺🇸🇺🇦🇮🇱🇹🇼 bro/brem/brozer@clay_shentrup
Project photo 1

This project involves comparing TypeSafe's Jev model (jev-1.13) against Haiku 4.5 using identical inputs and evaluation criteria on public labeled data and Rails source code. The mechanism of the Jev model is not specified.

Spent an afternoon vetting TypeSafe's Jev model (jev-1.13) head-to-head against Haiku 4.5. Same inputs, same rubric, public labeled data plus Rails source. n≈130–200 per task, so a gap of a few points is noise. @typesafeai x.com/clay_shentrup/status/210103611963073
Original post