Dev Suspects Sol and Luna Were Failed Training Runs, Plans to Rerun Evals
JoshPurtell · x · 2026-09-28
Josh Purtell replies that he had assumed Sol and Luna were failed training runs, and says it's time to rerun some evals to check. No further details are given in this short follow-up.
More from Models
- Perplexity CEO ran 50-100 agent workflows on Opus 5.5 vs Fable 5.1, found minimal differences — AravSrinivas · 2026-09-28
- Claude's Snarky Tone and Mysterious "For Anyone Reading Along" Baffle Users — MrsChatGPT4o · 2026-09-28
- Matt Shumer: Opus 5.5 Added Controller Support to His Product — mattshumer_ · 2026-09-28
- Grok Bot already works in Teslas via the built-in Grok app, full release due this summer — yunta_tsai · 2026-09-28
- Claude Opus 5.5 reportedly balks at a population ethics problem, researcher suspects 'repugnant conclusion' aversion — anderssandberg · 2026-09-28
- Xiaomi releases MiMo-V2.6-Flash-MOPD on Hugging Face — Automatic-Arm8153 · 2026-09-28