GPT-6 Astra beats Nethack on third try, sparking training-data transparency concerns
PMinervini · x · 2026-09-27
Ethan Mollick calls GPT-6 Astra beating Nethack on its third attempt a "startling achievement" — Nethack is the original roguelike and one of the hardest games ever; Mollick himself has never ascended.
Researcher pfau raises the key concern: nobody knows whether GPT-6 was specifically post-trained on Nethack. With zero transparency into training data and environments, the significance of such results cannot be independently evaluated.
More from Models
- Puppy Kill Bench: most models refuse, GPT6-Luna just executes the kill tool — MetroidsSuffering · 2026-09-27
- Ethan Mollick: Opus 4.7-5 lost the 'Claude feel', Opus 5.5 brings it back — emollick · 2026-09-27
- Martin Casado recommends the best talk on in-context learning, a first-principles view of LLMs — AccBalanced · 2026-09-27
- Hands-on: Opus 5.5 high beats GPT-6 astra xhigh on real Pagespeed optimization — mazzaTalk · 2026-09-27
- Dev on Opus 5.5: smooth multi-part coordination flips coding dynamic — ezshine · 2026-09-27
- Supersonic Labs open-sources Julia 1, a 144M-parameter CPU-runnable decision model — ThePrimeClock · 2026-09-27