July 2024 Summaries

The Align AI Research Review discusses a novel approach to reducing hallucinations in Large Language Models (LLMs) through Dynamic Memory Expert Selection. Li et al. introduced Lamini-1, a new model architecture that uses memory experts to store and dynamically retrieve facts. This research aims to minimize hallucinations, which can reduce from 50% to 5%. The Lamini Memory Tuning method enhances factual accuracy and reduces hallucinations by optimizing for zero error on specific facts rather than average error on all items. It also maintains the LLM's capacity to generalize while eliminating hallucinations regarding the information of interest. This research challenges the current consensus regarding the generalizability of LLMs and their capacity to generalize without hallucinations, emphasizing the importance of developing novel metrics and methodologies to assess the precision with which LLMs can memorize and recall facts.

Caltech's new study investigates how effectively designed input prompts can significantly impact large language model (LLM) outcomes, changing unlikely predictions into likely ones. The research conceptualizes LLMs as discrete stochastic dynamical systems and uses control theory to understand and modify their outputs. Prompt engineering is shown to have a major impact on LLM behavior. Limitations in existing work include the reliance on heuristics for prompt optimization, dependence on gradient information at the token embedding layer, and restricted analysis of LLM controllability to 'meaningful sentences.' The proposed system formalizes LLMs as a type of discrete stochastic dynamical system and analyzes the reachable set of system outputs. Empirical findings indicate that short prompt sequences can significantly change the chance of specific outputs, even transforming the least likely tokens into the most likely ones.