Meta- (out-of-context) learning in neural networks
Lists
Only you see this
Abstract
Brown et al. (2020) famously introduced the phenomenon of in-context learning in large language models (LLMs). We establish the existence of a phenomenon we call **meta-out-of-context learning (meta-OCL)** via carefully designed synthetic experiments with LLMs. Our results suggest that meta-OCL leads LLMs to more readily “internalize” the semantic content of text that is, *or appears to be*, broadly useful (such as true statements, or text from authoritative sources) and use it in appropriate circumstances. We further demonstrate meta-OCL in a synthetic computer vision setting, and propose two hypotheses for the emergence of meta-OCL: one relying on the way models store knowledge in their parameters, and another suggesting that the implicit gradient alignment bias of gradient-descent-based optimizers may be responsible. Finally, we reflect on what our results might imply about capabilities of future AI systems, and discuss potential risks. Our code is available at https://github.com/krasheninnikov/internalization.
BibTeX
@inproceedings{krasheninnikov2024meta,
author = {Dmitrii Krasheninnikov and Egor Krasheninnikov and Bruno Mlodozeniec and David Krueger},
title = {{Meta- (out-of-context) learning in neural networks}},
booktitle = {International Conference on Machine Learning},
year = {2024},
url = {https://openreview.net/forum?id=Fzp1DRzCIN},
}