• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
AI
Reaction

GPT-6 Astra reportedly beats NetHack on its third try

The player sharing the run called it a “startling achievement,” saying they have played a lot of NetHack but never won themselves.

Tim RocktäschelTR
Edward GrefenstetteEG
Dustin TranDT
25 Sources, 15d ago, first seen 15d ago

TLDR

A user says GPT-6 Astra beat NetHack on its third try. The linked write-up describes it as the first recorded NetHack win by a large language model agent, to its author's knowledge.

Combined views

427.2K

25 Sources, first seen 15d ago

3.2K likes140 comments860 saves249 reposts

Combined views

427.2K

25 Sources, first seen 15d ago

3.2K likes140 comments860 saves249 reposts

Sentiment

Positive63%37%Negative

Summary

Many accounts welcomed Astra's NetHack ascensions as an impressive benchmark for long-horizon reasoning and self-improvement, while others called the AGI framing overhyped or unnecessary.

Based on 86 sentiment-bearing replies from 79 accounts across 10 conversations.

Featured Source

Sentiment

Positive63%37%Negative

Summary

Many accounts welcomed Astra's NetHack ascensions as an impressive benchmark for long-horizon reasoning and self-improvement, while others called the AGI framing overhyped or unnecessary.

Based on 86 sentiment-bearing replies from 79 accounts across 10 conversations.

25 Sources

Edward Grefenstette@egrefenHot damn! Congrats on the AGI, @OpenAI! 😉 https://kenforthewin.github.io/blog/posts/llm-nethack-ascension/15d
heiner@HeinrichKuttlerLooks like it may be AGI now. https://kenforthewin.github.io/blog/posts/llm-nethack-ascension15d
Ethan Mollick@emollickIn all seriousness, this is a startling achievement for GPT-6 Astra. https://kenforthewin.github.io/blog/posts/llm-nethack-ascension/#run=astra-3&frame=0&turn=1 (This is GPT-6 Astra beating Nethack on its 3rd try. Nethack is the original roguelike and one of the most famously hard games of all time. I have played a lot, and I've never ascended)15d
Danielle Fong 🔆@DanielleFongWow!!! Nethack!! This is actually a huge milestone15d
Tim Rocktäschel@_rocktAaaaaaaand it's apparently solved 🫣😅15d
Machine Learning Street Talk@MLStreetTalkRT @egrefen: Hot damn! Congrats on the AGI, @OpenAI! 😉 https://kenforthewin.github.io/blog/posts/llm-nethack-ascension/15d
David Pfau@pfauSurprised this isn't a bigger deal.14d
Dustin Tran@dustinvtrani remember the nethack challenge at neurips. seriously very challenging (and a great game). it is perhaps a testament to llms' ability to self-improve their harness and plan and take action over long horizons. llm's multimodal reasoning still remains a bottleneck for modern games - it will take longer to beat 3d games where camera control is a basic mechanic.14d
Steven Hansen@ZergylordNetHack is a better AGI eval than ARC-AGI.14d
Jeff Clune@jeffcluneReally surprising how quickly it went from not working for years, to just barely working, to solved. That's pretty much the story for every domain AI is applied to. Once it barely starts working, it will be superhuman very soon. RSI is just barely working now.14d
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI
    GPT-6 Astra

    25 Sources

    Edward Grefenstette@egrefenHot damn! Congrats on the AGI, @OpenAI! 😉 https://kenforthewin.github.io/blog/posts/llm-nethack-ascension/15d
    heiner@HeinrichKuttlerLooks like it may be AGI now. https://kenforthewin.github.io/blog/posts/llm-nethack-ascension15d
    Ethan Mollick@emollickIn all seriousness, this is a startling achievement for GPT-6 Astra. https://kenforthewin.github.io/blog/posts/llm-nethack-ascension/#run=astra-3&frame=0&turn=1 (This is GPT-6 Astra beating Nethack on its 3rd try. Nethack is the original roguelike and one of the most famously hard games of all time. I have played a lot, and I've never ascended)15d
    Danielle Fong 🔆@DanielleFongWow!!! Nethack!! This is actually a huge milestone15d
    Tim Rocktäschel@_rocktAaaaaaaand it's apparently solved 🫣😅15d
    Machine Learning Street Talk@MLStreetTalkRT @egrefen: Hot damn! Congrats on the AGI, @OpenAI! 😉 https://kenforthewin.github.io/blog/posts/llm-nethack-ascension/15d
    David Pfau@pfauSurprised this isn't a bigger deal.14d
    Dustin Tran@dustinvtrani remember the nethack challenge at neurips. seriously very challenging (and a great game). it is perhaps a testament to llms' ability to self-improve their harness and plan and take action over long horizons. llm's multimodal reasoning still remains a bottleneck for modern games - it will take longer to beat 3d games where camera control is a basic mechanic.14d
    Steven Hansen@ZergylordNetHack is a better AGI eval than ARC-AGI.14d
    Jeff Clune@jeffcluneReally surprising how quickly it went from not working for years, to just barely working, to solved. That's pretty much the story for every domain AI is applied to. Once it barely starts working, it will be superhuman very soon. RSI is just barely working now.14d
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet