Stanford AI Lab(@StanfordAILab)
What would pretraining with zero real data look like? Can a randomly initialized model learn to gene...
6.4内容质量
TL;DR · AI 摘要
Stanford AI Lab on X: "What would pretraining with zero real data look like? Can a randomly initialized model learn to g...
核心要点
- 主题聚焦：What would pretraining with zero real data look
- 来源：Stanford AI Lab(@StanfordAILab)，建议结合原文判断细节。
- AI 分析暂不可用，本条为保底评分与摘要。
#AI
打开原文Stanford AI Lab on X: "What would pretraining with zero real data look like? Can a randomly initialized model learn to generate all of its training data entirely through self-play? Find out below! :)" / X
Stanford AI Lab
@StanfordAILab
What would pretraining with zero real data look like? Can a randomly initialized model learn to generate all of its training data entirely through self-play? Find out below! :)
Michael Y. Li
@michaelyli_
Sep 25
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
Show more
5:58 PM · Sep 25, 2026
·
43.7K
Views
13
31
354
246