An official website of the United States government
Neural models, including large language models (LLMs), achieve superior performance on logical reasoning tasks such as question answering. To elicit reasoning capabilities from LLMs, recent works propose using the chain-of-thought (CoT) mechanism to generate both the reasoning chain and the answer, which enhances the model’s capabilities in conducting reasoning. However, due to LLM’s uninterpretable nature and the extreme flexibility of free-form explanations, several challenges remain: such as struggling with inaccurate reasoning, hallucinations, and not aligning with human preferences. In this talk, we will focus on (1) our design of leveraging structured information (that is grounded to the context), for the explainable complex question answering and reasoning; (2) our multi-module interpretable framework for inductive reasoning, which conducts step-wise faithful reasoning with iterative feedback.
more »
« less
Language Models Don’t Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Prompting
Large Language Models (LLMs) can achieve strong performance on many tasks by producing step-by-step reasoning before giving a final output, often referred to as chain-of-thought reasoning (CoT). It is tempting to interpret these CoT explanations as the LLM’s process for solving a task. This level of transparency into LLMs’ predictions would yield significant safety benefits. However, we find that CoT explanations can systematically misrepresent the true reason for a model’s prediction. We demonstrate that CoT explanations can be heavily influenced by adding biasing features to model inputs—e.g., by reordering the multiple-choice options in a few-shot prompt to make the answer always “(A)”—which models systematically fail to mention in their explanations. When we bias models toward incorrect answers, they frequently generate CoT explanations rationalizing those answers. This causes accuracy to drop by as much as 36% on a suite of 13 tasks from BIG-Bench Hard, when testing with GPT-3.5 from OpenAI and Claude 1.0 from Anthropic. On a social-bias task, model explanations justify giving answers in line with stereotypes without mentioning the influence of these social biases. Our findings indicate that CoT explanations can be plausible yet misleading, which risks increasing our trust in LLMs without guaranteeing their safety. Building more transparent and explainable systems will require either improving CoT faithfulness through targeted efforts or abandoning CoT in favor of alternative methods.
more »
« less
- Award ID(s):
- 2046556
- PAR ID:
- 10542779
- Publisher / Repository:
- MIT Press
- Date Published:
- Journal Name:
- Advances in neural information processing systems
- Volume:
- 36
- ISSN:
- 1049-5258
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Large language models (LLMs) have demonstrated an impressive ability to perform arithmetic and symbolic reasoning tasks, when provided with a few examples at test time ("few-shot prompting"). Much of this success can be attributed to prompting methods such as "chain-of-thought", which employ LLMs for both understanding the problem description by decomposing it into steps, as well as solving each step of the problem. While LLMs seem to be adept at this sort of step-by-step decomposition, LLMs often make logical and arithmetic mistakes in the solution part, even when the problem is decomposed correctly. In this paper, we present Program-Aided Language models (PAL): a novel approach that uses the LLM to read natural language problems and generate programs as the intermediate reasoning steps, but offloads the solution step to a runtime such as a Python interpreter. With PAL, decomposing the natural language problem into runnable steps remains the only learning task for the LLM, while solving is delegated to the interpreter. We demonstrate this synergy between a neural LLM and a symbolic interpreter across 13 mathematical, symbolic, and algorithmic reasoning tasks from BIG-Bench Hard and others. In all these natural language reasoning tasks, generating code using an LLM and reasoning using a Python interpreter leads to more accurate results than much larger models. For example, PAL using Codex achieves state-of-the-art few-shot accuracy on GSM8K, surpassing PaLM which uses chain-of-thought by absolute 15% top-1.more » « less
-
This study aimed to demonstrate the effectiveness of a chain-of-thought (CoT) prompt tailored to generate personalized energy-saving strategies for households. Energy eco-systems grow increasingly complex on both demand and supply sides, but households often lack clear guidance on how to adjust their energy usage patterns in response to dynamic pricing structures and variable consumption patterns. While recent studies have indicated the potential of integrating large language models (LLMs) into home energy management (HEM) tasks, these efforts have primarily focused on technical and analytical applications, with limited attention to human-centered approaches that directly support occupants in managing their energy use. To address this gap, this study systematically assessed the performance of an LLM-integrated HEM system (HEMS) in generating personalized, data-driven, and behavior-oriented energy-saving recommendations for an actual household located in Austin, Texas. The proposed CoT prompt, incorporating advanced analytical and step-by-step reasoning processes - was compared against a concise, dialogue-type prompt to evaluate the effectiveness of structured reasoning in enhancing the accuracy and contextual relevance of LLM-generated recommendations. Results show that the CoT prompt achieved a mean accuracy of 71.3%, significantly outperforming the concise, dialogue-type prompt (29.3%). It indicated that structured reasoning enabled the LLM to generate appliance specific, data-driven, and context-aware interventions rather than generic, surface-level solutions. This study advances the understanding of how LLM-integrated HEMS can foster more intuitive, informed, and data-driven household energy decisions.more » « less
-
Large language models (LLMs) excel in complex reasoning tasks, and distilling their reasoning capabilities into smaller models has shown promise. However, we uncover an interesting phenomenon, which we term the Small Model Learnability Gap: small models (3B parameters) do not consistently benefit from long chain-of-thought (CoT) reasoning or distillation from larger models. Instead, they perform better when fine-tuned on shorter, simpler reasoning chains that better align with their intrinsic learning capacity. To address this, we propose Mix Distillation, a simple yet effective strategy that balances reasoning complexity by combining long and short CoT examples or reasoning from both larger and smaller models. Our experiments demonstrate that Mix Distillation significantly improves small model reasoning performance compared to training on either data alone. These findings highlight the limitations of direct strong model distillation and underscore the importance of adapting reasoning complexity for effective reasoning capability transfer.more » « less
-
ProtoVQA: An Adaptable Prototypical Framework for Explainable Fine-Grained Visual Question AnsweringVisual Question Answering (VQA) is increasingly used in diverse applications ranging from general visual reasoning to safety-critical domains such as medical imaging and autonomous systems, where models must provide not only accurate answers but also explanations that humans can easily understand and verify. Prototype-based modeling has shown promise for interpretability by grounding predictions in semantically meaningful regions for purely visual reasoning tasks, yet remains underexplored in the context of VQA. We present ProtoVQA, a unified prototypical framework that (i) learns question-aware prototypes that serve as reasoning anchors, connecting answers to discriminative image regions, (ii) applies spatially constrained matching to ensure that the selected evidence is coherent and semantically relevant, and (iii) supports both answering and grounding tasks through a shared prototype backbone. To assess explanation quality, we propose the Visual–Linguistic Alignment Score (VLAS), which measures how well the model’s attended regions align with ground-truth evidence. Experiments on Visual7W show that ProtoVQA yields faithful, fine-grained explanations while maintaining competitive accuracy, advancing the development of transparent and trustworthy VQA systems.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript
- Journal Article:
- The DOI is not currently available.
