Tim Rocktäschel

@_rockt

Teaching AI the joy of invention at @Recursive_SI, Professor of AI @AI_UCL, PI @BOLD_Lab_AI, Fellow @ELLISforEurope. Ex @GoogleDeepMind @AIatMeta @CompSciOxford

London, England

rockt.ai

Replies Reposts Media

Pinned
Tim Rocktäschel
@_rockt
May 28, 2025
My @iclr_conf 2025 Keynote on "Open-Endedness, World Models, and the Automation of Innovation" is now publicly available: iclr.cc/virtual/2025/i…
13
Tim Rocktäschel
@_rockt
Sep 25
Impressive. Luckily we can turn @NetHack_LE / balrogai.com into an ASI benchmark in an instant. The human world record is 61 consecutive NetHack ascensions (reddit.com/r/nethack/comm…). Also curious to see if Astra can ascend in all roles (nethackwiki.com/wiki/Z-score).
Edward Grefenstette
@egrefen
Sep 25
Hot damn! Congrats on the AGI, @OpenAI! 😉 kenforthewin.github.io/blog/posts/llm…
9
Tim Rocktäschel
@_rockt
Sep 25
Aaaaaaaand it's apparently solved 🫣😅
Tim Rocktäschel
@_rockt
Sep 21
The NetHack Learning Environment (@NetHack_LE) has been around since 2020. Since 2024, balrogai.com is evaluating frontier LLM performance on it. We are slowly seeing progress on it, but it still seems far from being solved.
19
Tim Rocktäschel
@_rockt
Sep 21
The NetHack Learning Environment (@NetHack_LE) has been around since 2020. Since 2024, balrogai.com is evaluating frontier LLM performance on it. We are slowly seeing progress on it, but it still seems far from being solved.
7
Tim Rocktäschel
@_rockt
Sep 19
Great progress. Not yet AGI though.
Davide Paglieri
@PaglieriDavide
Sep 19
BALROG’s leaderboard has three new entries, courtesy of @creus_roger Very interesting that GPT 5.6 Sol at max effort is still within error of Gemini 3 and 3.1 pro. GPT Astra 6 on max reasoning effort however reaches new heights, and a @NetHack_LE avg. progression of 13% 🏰
3