Cognitive Training for Language Models: Towards General Capabilities via Cross-Entropy Games

TL;DR

This paper introduces a framework called cognitive training, using Cross-Entropy Games to automatically build curricula that enhance language models' general capabilities through skill discovery.

Contribution

It proposes a universal family of tasks and a greedy optimization approach for curriculum growth, offering a principled method for skill discovery in language models.

Findings

The framework formalizes curriculum growth via a meta-objective.

Under certain assumptions, the process is essentially unique.

Cognitive training can potentially lead to general capabilities in language models.

Abstract

Defining a constructive process to build general capabilities for language models in an automatic manner is considered an open problem in artificial intelligence. Towards this, we consider the problem of building a curriculum of tasks that grows a model via relevant skill discovery. We provide a concrete framework for this task, using a family of tasks called Cross-Entropy Games, which we postulate is universal in a suitable sense. We show that if it is possible to grow the curriculum for relevant skill discovery by iterating a greedy optimization algorithm, then, under natural assumptions, there is essentially only one meta-objective possible (up to a few hyper-parameters). We call the resulting process cognitive training. We postulate that, given sufficiently capable language models as players and meta-samplers, cognitive training provides a principled way to relevant skill discovery; and hence to the extent general capabilities are achievable via greedy curriculum learning, cognitive training would be a solution.