A 2B Model Trained for Funsearch with Shell Tools, Heading into a Month-Long Autonomous Run
cephaloform · x · 2026-11-21
The author trained a 2B model specifically for funsearch—an old autoresearch harness that even works on base models—but this time with tools (shell). Plan: let it run for the rest of the day, set up a large batch of environments tomorrow, debug the text-to-wav step (maybe needs an example), wildly increase batch size, delete broken envs, then kick off a month-long run.
Related event: Developer Trains Dedicated 2B Model to Run FunSearch with Shell Access(2 posts)→
More from Research
- Ex-NVIDIA intern shares world-model memory research built on Matrix-Game 2.0 — cindy_x_wu · 2026-11-21
- METR's Time Horizon 2.0 report ships too few appendices, researchers complain — leaked topic list — dfrsrchtwts · 2026-11-21
- Proposed proof of Generalized Riemann Hypothesis released; mathematician stresses it is entirely human, no AI — paulabartabajo_ · 2026-11-21
- Arcfront turns real hospital and factory work records into RL environments and frontier model evals — rohanpaul_ai · 2026-11-21
- Goodfire probes Inkling 2 Mini: looped model layers read from "future" layers — fatihdin4en · 2026-11-21
- Terence Tao offers a bet to anyone who really believes RH will be proved next year — jaivin · 2026-11-21