Dev RL-tunes open-weight Inkling 2 Small on Celeste, +65% on self-made bench
simonguozirui · x · 2026-11-13
A developer built "Modeline 584B A17B" on top of Thinking Machines' newly open-weight Inkling 2 Small, using the Tinker API to run RL on procedurally generated new levels of Celeste Classic. The model improved +65% on the author's own CelesteBench, and its chains of thought are reportedly fun to read. A playful demo of RL on game environments.
More from Fun
- 'Hear me out, Jensen shouldn't have been fired' — an AI-circle meme take — vxnuaj · 2026-11-13
- "Singapore" reportedly accounts for 15-20% of NVIDIA revenue, and the jokes write themselves — JensHonack · 2026-11-13
- Joke "warez abliterated models" post name-drops unreleased GLM 5.3 and DeepSeek V4.1 — HankYeomans · 2026-11-13
- Pickover revisits "Apocalypse Numbers": the 3184th Fibonacci number has 666 digits — pickover · 2026-11-13
- Viral meme imagines Jensen Huang volunteering as the AI bubble's sin eater — minhnguyen · 2026-11-13
- AI creator receives death threats by SMS, exposing toxicity inside AI community — AIandDesign · 2026-11-13