Understanding In-Context Learning in Transformers and LLMs by Learning to Learn Discrete Functions

TL;DR

This paper investigates how Transformers and large language models learn algorithms for Boolean functions, revealing their capabilities, limitations, and adaptability in in-context learning scenarios through various experiments.

Contribution

It demonstrates the extent of Transformers' ability to learn different algorithms, compares attention-based and attention-free models, and evaluates LLMs' performance on unseen tasks.

Findings

Transformers perform well on simple tasks but struggle with complex ones.

Attention-free models show similar capabilities to Transformers.

LLMs like GPT-4 can effectively predict unseen data, rivaling nearest-neighbor methods.

Abstract

In order to understand the in-context learning phenomenon, recent works have adopted a stylized experimental framework and demonstrated that Transformers can learn gradient-based learning algorithms for various classes of real-valued functions. However, the limitations of Transformers in implementing learning algorithms, and their ability to learn other forms of algorithms are not well understood. Additionally, the degree to which these capabilities are confined to attention-based models is unclear. Furthermore, it remains to be seen whether the insights derived from these stylized settings can be extrapolated to pretrained Large Language Models (LLMs). In this work, we take a step towards answering these questions by demonstrating the following: (a) On a test-bed with a variety of Boolean function classes, we find that Transformers can nearly match the optimal learning algorithm for 'simpler' tasks, while their performance deteriorates on more 'complex' tasks. Additionally, we find that certain attention-free models perform (almost) identically to Transformers on a range of tasks. (b) When provided a teaching sequence, i.e. a set of examples that uniquely identifies a function in a class, we show that Transformers learn more sample-efficiently. Interestingly, our results show that Transformers can learn to implement two distinct algorithms to solve a single task, and can adaptively select the more sample-efficient algorithm depending on the sequence of in-context examples. (c) Lastly, we show that extant LLMs, e.g. LLaMA-2, GPT-4, can compete with nearest-neighbor baselines on prediction tasks that are guaranteed to not be in their training set.