lamss: when large language models meet self-skepticism

2024-09-10

AI Summary

Large language models (LLMs) suffer from hallucination due to insufficient self-awareness and inability to assess their own confidence. Method: We propose a learnable self-doubt mechanism that explicitly models human skeptical reasoning as a discrete, decodable multi-level doubt token sequence. This is realized via semantic-enhanced vocabulary expansion, coupled with joint pretraining and fine-tuning, doubt-degree–conditioned decoding, and threshold-based response gating—enabling fine-grained confidence estimation and dynamic response rejection. Contribution/Results: To our knowledge, this is the first work to explicitly encode doubt capability as trainable tokens. Evaluated on multiple-choice and open-domain question answering, our approach significantly improves accuracy, AUC, and average precision (AP), while demonstrating strong generalization across tasks and robustness across domains.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Hallucination is a major challenge for large language models (LLMs), prevent ing their further application in some fields. The skeptical thinking of humankind could be useful for LLMs to self-cognition, self-reflection and alleviate their hal lucinations. Inspired by this consideration, we propose a novel approach called LaMsS, which combines the semantic understanding capability of LLMs with self-skepticism. By introducing a series of skepticism tokens and augmenting them into the vocabulary, we conduct both pertaining and finetuning, which allow the LLM to decode each normal token followed by a skeptical token, represent ing different skepticism levels. By calculating the response skepticism given a query, one can define a new self-aware LLM which is only willing to answer with relative lower skepticism level than the threshold. By examining the accu racy, AUC and AP of willingly answering questions, we demonstrate that LaMsS achieves better performance than baselines on both multi-choice questions and open-domain question-answering benchmarks, and can generalize to multi-task and out-of-domain settings. Our study sheds some lights on the self-skepticism modeling on further artificial intelligence. Project code and model checkpoints can be found in https://anonymous.4open.science/r/SM-1E76.

Problem

• Reducing hallucinations in large language models (LLMs).
• Enhancing LLMs with self-skepticism for better self-awareness.
• Improving accuracy and reliability in multi-task question-answering.

Innovation

• Introduces skepticism tokens for self-reflection
• Augments vocabulary for multi-level skepticism
• Threshold-based response filtering for accuracy