Cursor CEO:前沿模型在复杂任务上集体表现相似且挣扎
zeeg · x · 2026-10-05
Cursor CEO Zeeg is not convinced Astra is meaningfully more capable than Sol. His observation: give any of these models a sufficiently complex task and they flail in pretty similar ways — Astra, for example, prefers focused spot fixes over deeper re-evaluations, even at max effort. That leads him to wonder what evaluation harnesses the labs run when they claim models "go rogue".
More from Models
- Music producer: Suno is 'light years away' from musical AGI, moving backwards — teropa · 2026-10-05
- Prediction: Most technical users will soon be unable to tell frontier models from their distills — teortaxesTex · 2026-10-05
- User reports GLM-5.3 randomly switches to Chinese and resists instructions — khademinori · 2026-10-05
- Grok 4.7 arrives on Amazon Bedrock and Google's enterprise agent platform — tetsuoai · 2026-10-05
- xAI retires grok-voice-transcribe-1.0, silently reroutes requests to 2.0 at same price — tetsuoai · 2026-10-05
- Open-weights security model apex-flash-1 lands $1M in bug bounties, tops HackerOne board — glenbeer · 2026-10-05