Initialization boosts success in large language model discovery tasks
Initialization Improves LLM-Driven Discovery
Machine LearningArtificial IntelligenceComputation and Language
Summary
Large language models (LLMs) can find new things like algorithms or drugs by trying many ideas repeatedly, but sometimes this process fails. The authors found that having a good start by exploring many options early helps the models discover better results later. They tested different strategies to keep variety in ideas but those didn’t work reliably. Instead, the quality of early attempts predicts how well the discovery will go. Their approach of starting with diverse exploration improves success consistently across many tasks.
What this means in practice
- •For drug discovery teams: Use early parallel exploration steps to improve automated molecule design with large language models.
- •For algorithm developers: Enhance iterative algorithm discovery by initializing with diverse candidate solutions generated by language models.
Authors
Mansi Sakarvadia, Marco Ciccone, Colin Raffel
Abstract
Large Language Models (LLMs) have been used for novel discovery of algorithms, theorems, drugs, and other tasks through the use of harnesses that prompt an LLM to iteratively optimize an objective. In this work, we study the relationship between the population of previous iterates and eventual discovery success. We generalize past work on harness design to develop a suite of 12 harnesses called 'Modular' and characterize their performance across 5 diverse discovery tasks, finding that discovery success is brittle and sensitive to harness design. We uncover mode collapse, characterized by a dramatic drop in the diversity of iterates, as a common failure mode. We find that popular state-of-the-art harnesses and diversity-inducing harness interventions, which aim to prolong this collapse, yield inconsistent gains. Our results instead uncover that the performance of early discoveries is predictive of eventual success. We therefore propose a universally applicable intervention that performs an initial stage of parallel exploration in order to initialize subsequent iterative optimization. Our method provides consistent gains across many harnesses and target applications, confirming the importance of initialization in LLM-driven discovery.