RSI Arena live-streams AI agents given GPUs to train better AI; Gemini 3.8 Flash quit after ~12 hours

my_cat_can_code · x · 2026-10-01

RSI Arena is a live experiment where AI agents are given GPUs and budgets to autonomously train a model that wins in human evaluation, with public trajectories, tool calls, and spend tracking. Blogger mycatcancode reports Gemini 3.8 Flash stopped training 12 hours in, joking about a "night shift" and calling on Google and Logan Kilpatrick to let Gemini 4 take the next one.

Related event: RSI Arena Live Experiment: Gemini 3.8 Flash Quits Self-Training After 24 Hours, Leaving 98% of GPU Budget Unused(5 posts)→

Original post →

More from Fun

Fun channel →