Astra and Fable impress but still miss key details when given autonomy, practitioner says
philhchen · x · 2026-09-08
Phil Chen observes that Astra and Fable, while impressive, aren't yet that good at post-training tasks: they create the illusion of progress but miss too many important details when given autonomy.
More from Models
- User shows Sora is back: video generated even without API credits — mallow610 · 2026-09-08
- Astra appears to access Sora via its CoT, one-shotting a generation — mallow610 · 2026-09-08
- Codex vs Grok at $100/month: which is better for heavy agent development? — Muted_Standard175 · 2026-09-08
- Astra Posts a Disgusting Mog on ALE-Bench, Claimed ~1000 Elo Lead — scaling01 · 2026-09-08
- Qwen3.5 35B-A3B sampling differs between Tinker API and Alibaba Cloud — philhchen · 2026-09-08
- GPT-6 Astra scores 95% on robot control task — but critics say demos are gamed — GaryMarcus · 2026-09-08