Learning to Plan Long-Term for Language Modeling

TL;DR

This paper introduces a planning mechanism for language models that predicts long-term latent plans, enabling better future text prediction by trading off computation time for accuracy.

Contribution

It proposes a novel latent planning approach that improves long-term coherence and prediction accuracy in language models.

Findings

Enhanced next token prediction accuracy

Ability to trade computation time for better predictions

Improved long-term coherence in generated text

Abstract

Modern language models predict the next token in the sequence by considering the past text through a powerful function such as attention. However, language models have no explicit mechanism that allows them to spend computation time for planning long-distance future text, leading to a suboptimal token prediction. In this paper, we propose a planner that predicts a latent plan for many sentences into the future. By sampling multiple plans at once, we condition the language model on an accurate approximation of the distribution of text continuations, which leads to better next token prediction accuracy. In effect, this allows trading computation time for prediction accuracy.