49.7% of fitness-improving mutations depended on prior neutral changes in digital-life experiment
bushibuilds · x · 2026-10-10
Core result of the E2 experiment: each time a mutation made a program run longer, the team checked whether that exact mutation would have worked before the preceding neutral changes. Of 6,869 improvements, 3,417 (49.7%) would not have — silent changes altered what was possible next.
The walks also surfaced behaviors E1 rarely or never saw: 86% returned data, 94% reached persistent storage writes, 96% executed CREATE, 69% executed CALL-family instructions, 14% found loops, and some called themselves. None reproduced.
More from Research
- Inferring goals from failure: online Bayesian goal inference for boundedly-rational agents — xuanalogue · 2026-10-11
- Alignment researcher points to Rohin Shah's value learning sequence and IRL model misspecification — xuanalogue · 2026-10-11
- Looped LM paper: 1.6B model matches full-cache baseline with 3x smaller KV cache — rupspace · 2026-10-11
- Pure RL discovers superhuman robot strategies in sim, transfers zero-shot to real hardware — KyleMorgenstein · 2026-10-11
- Softmax picks probabilities, cross-entropy picks the target: a 3-class walkthrough — techNmak · 2026-10-11
- PartLLM brings LLM-powered 3D mesh part segmentation to SIGGRAPH Asia with code released — Promptmethus · 2026-10-11