English

Transcendence: Generative Models Can Outperform The Experts That Train Them

Machine Learning 2024-10-15 v4 Artificial Intelligence

Abstract

Generative models are trained with the simple objective of imitating the conditional probability distribution induced by the data they are trained on. Therefore, when trained on data generated by humans, we may not expect the artificial model to outperform the humans on their original objectives. In this work, we study the phenomenon of transcendence: when a generative model achieves capabilities that surpass the abilities of the experts generating its data. We demonstrate transcendence by training an autoregressive transformer to play chess from game transcripts, and show that the trained model can sometimes achieve better performance than all players in the dataset. We theoretically prove that transcendence can be enabled by low-temperature sampling, and rigorously assess this claim experimentally. Finally, we discuss other sources of transcendence, laying the groundwork for future investigation of this phenomenon in a broader setting.

Keywords

Cite

@article{arxiv.2406.11741,
  title  = {Transcendence: Generative Models Can Outperform The Experts That Train Them},
  author = {Edwin Zhang and Vincent Zhu and Naomi Saphra and Anat Kleiman and Benjamin L. Edelman and Milind Tambe and Sham M. Kakade and Eran Malach},
  journal= {arXiv preprint arXiv:2406.11741},
  year   = {2024}
}

Comments

Code, models, and data at https://transcendence.eddie.win

R2 v1 2026-06-28T17:08:57.571Z