@stanfordnlp: This is cool! We’ll have grown up LLMs nurturing baby LLMs to learn by self-play in the crib in no time! More seriously…
Summary
The article introduces Self-Play Pretraining with Zero Data, a research method where two LLMs start from random initialization and learn through a generator proposing programs for a universal Turing machine, demonstrating how inductive biases can enable general learning.
View Cached Full Text
Cached at: 09/25/26, 10:47 PM
This is cool! We’ll have grown up LLMs nurturing baby LLMs to learn by self-play in the crib in no time!
More seriously, it’s a good demonstration of how general inductive biases can be an effective “Universal Grammar” for more general learning about the world.
Michael Y. Li (@michaelyli_): Can an LM, starting from random init (!!), learn to generate all of its pretraining data?
Introducing 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 their
Similar Articles
@yoheinakajima: another thing to read later to see if i can understand
The tweet introduces a research concept called Self-Play Pretraining with Zero Data, where two models start from random initialization and generate pretraining data through self-play on a universal Turing machine.
G-Zero: Self-Play for Open-Ended Generation from Zero Data
This paper introduces G-Zero, a verifier-free framework that enables autonomous large language model self-improvement through co-evolutionary training using intrinsic rewards and hint-based guidance. It aims to overcome the limitations of proxy LLM judges in open-ended tasks by deriving supervision from internal distributional dynamics.
Self-play helped AI achieve superhuman performance in Go, so why hasn’t it done the same for LLMs? Researchers have found a solution.
Researchers introduce Self-Guided Self-Play (SGS), a self-play algorithm for LLMs that prevents reward hacking by using a Guide role to score synthetic problems. Applied to theorem proving in Lean4, SGS surpasses RL baselines and allows a 7B model to outperform a 671B model.
PopuLoRA: Co-Evolving LLM Populations for Reasoning Self-Play
PopuLoRA introduces a population-based asymmetric self-play framework for RLVR post-training of LLMs, where teacher and student LoRA adapters co-evolve to generate increasingly complex problems, overcoming the self-calibration limitation of single-agent self-play.
LLMZero: Discovering Adaptive Training Strategies for RL Post-Training via LLM Agents
LLMZero uses LLM agents to search over training trajectories via tree search, discovering adaptive multi-parameter transitions for RL post-training that outperform fixed schedules and grid search across diverse tasks.