A Taxonomy of Transcendence
It addresses the problem of understanding model capabilities beyond human-level performance for AI researchers, though it appears incremental by building on prior work.
The paper investigates how language models can outperform their training data sources by identifying data properties that enable transcendence, categorizing it into three modes: skill denoising, skill selection, and skill generalization.
Although language models are trained to mimic humans, the resulting systems display capabilities beyond the scope of any one person. To understand this phenomenon, we use a controlled setting to identify properties of the training data that lead a model to transcend the performance of its data sources. We build on previous work to outline three modes of transcendence, which we call skill denoising, skill selection, and skill generalization. We then introduce a knowledge graph-based setting in which simulated experts generate data based on their individual expertise. We highlight several aspects of data diversity that help to enable the model's transcendent capabilities. Additionally, our data generation setting offers a controlled testbed that we hope is valuable for future research in the area.