Researcher: GPT6 Astra brings a step change in review quality, guided prompts help further
anshulkundaje · x · 2026-10-29
Stanford researcher Anshul Kundaje reports that after regular testing, GPT6 Astra (Ultra) marks a step change in review quality. Guiding the model with a short prompt to clearly segregate fatal flaws, big-picture issues, technical issues, nitpicks and wishlist items improves output further. A replier notes this likely assumes fine-tuning on very good reviews.
Related event: Stanford Researcher Sees Step-Change in GPT6 Astra's Peer Review Quality(2 posts)→
More from Models
- Anthropic says no regression behind Claude Code usage complaints, blames heavy sessions — bcherny · 2026-10-29
- OpenAI: Codex and Chat capabilities will merge this year, unified limits feared — CtrlAltDwayne · 2026-10-29
- Community reverse-engineers Qwen3.5-27B from public weights: it's grown from the 9B model — davidad · 2026-10-29
- Reddit User: AI Labs' Tier Sprawl Is Broken — One Model, One Price Would Win — Tight-Grocery9053 · 2026-10-29
- Mystery model 'Kerria' spotted on LMArena, tokenizer errors suggest it could be Qwen's next model — lyraxana · 2026-10-29
- Google employee says Gemini's writing style is more pleasant to read than other LLM-speak — amankhan · 2026-10-29