Self-play pretraining with zero data: models bootstrap from random init
burny_tech · x · 2026-09-27
A proof-of-concept called Self-Play Pretraining with Zero Data: two models start from random initialization — a generator proposes programs for a universal Turing machine and a learner trains on the outputs — never touching real data. Zero-shot validation loss on images, text, audio, and melodies decreases predictably with self-play compute, and the learner develops in-context learning.
The sharer's framing: ML progress has always been bottlenecked by our courage to propose ideas that "can't possibly work" and test them anyway. Co-led by Stanford researchers with Noah Goodman and Yoav Levine among co-authors.
More from Research
- Explosion launches Ellf directory: 52 NLP templates plus multi-agent annotation workflow — _inesmontani · 2026-11-11
- STABLE Optimizer Accepted at NeurIPS 2026: Taming Continual Learning Forgetting at the Optimizer Level — v_lomonaco · 2026-11-11
- From Bahdanau to FlashAttention: a timeline of attention's 8-year evolution — cneuralnetwork · 2026-11-11
- Three years building CPUs and spiking nets: one engineer's path to AI accelerators — k7agar · 2026-11-11
- When AI becomes an oracle, what happens to mathematics? — stevenstrogatz · 2026-11-11
- 200 Years of Linear Algebra in One Chart: From Leibniz's Determinants to Peano's Vector Spaces — burny_tech · 2026-11-11