Faithful Reasoning Using Large Language Models

By Antonia Creswell, Murray Shanahan. 2022

163 citations recorded by the source corpus.

Although contemporary large language models (LMs) demonstrate impressive question-answering capabilities, their answers are typically the product of a single call to the model. This entails an unwelcome degree of opacity and compromises performance, especially on problems that are inherently multi-step. To address these limitations, we show how LMs can be made to perform faithful multi-step reasoning via a process whose causal structure mirrors the underlying logical structure of the problem. Our approach works by chaining together reasoning steps, where each step results from calls to two fine-tuned LMs, one for selection and one for inference, to produce a valid reasoning trace. Our method carries out a beam search through the space of reasoning traces to improve reasoning quality. We demonstrate the effectiveness of our model on multi-step logical deduction and scientific question-answering, showing that it outperforms baselines on final answer accuracy, and generates humanly interpretable reasoning traces whose validity can be checked by the user.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Natural Language Processing, Prompting & Reasoning

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Natural Language Processing

In Prompting & Reasoning

Citations in the corpus

ReAct: Synergizing Reasoning and Acting in Language Models (2022, neutral): “Faithful reasoning (Creswell & Shanahan, 2022) decomposes multi-step reasoning into three steps, each performed by a dedicated LM respectively.”
Augmented Language Models: a Survey (2023, background): “A way to produce faithful reasoning traces is to generate pairs of questions and their corresponding answers for each reasoning step (Creswell and Shanahan, 2022), but there is still no guarantee of the correctness of these intermediate steps.”
Tree of Thoughts: Deliberate Problem Solving with Large Language Models (2023, background): “See concrete examples of how thoughts are generated, evaluated, and searched in Figures 2,4,6. choices instead of just picking one, and (2) evaluates its current status and actively looks ahead or backtracks to make more global decisions.”
Reasoning with Language Model Prompting: A Survey (2022, unknown): “Other promising directions include retrieval augmentation (Li et al., 2022a), model editing (Cao et al., 2021; Mitchell et al., 2022a,b; Cheng et al., 2023), delta-tuning (He et al., 2022; Mao et al., 2022; Pal et al., 2022; Ding et al., 2022), etc. Robust, Faithful and Interpretable Reasoning.”
Towards Reasoning in Large Language Models: A Survey (2022, methodology): “For instance, Creswell et al. (2022); Creswell and Shanahan (2022) introduce a selection-inference framework that uses LLMs as modules to select and infer reasoning steps from a set of facts that culminate in the final answer.”
Contrastive Distillation Is a Sample-Efficient Self-Supervised Loss Policy for Transfer Learning (2022, unknown): “Creswell & Shanahan (2022) demonstrate a framework for faithful reasoning on EntailmentBank and Proofwriter (Dalvi et al., 2022; Tafjord et al., 2021).”
Teaching Algorithmic Reasoning via In-context Learning (2022, methodology): “Creswell and Shanahan (2022) uses a similar multi-model and multi-prompt strategy in order to separate out selection from inference in reasoning problems.”
Honest Students from Untrusted Teachers: Learning an Interpretable Question-Answering Pipeline from a Pretrained Language Model (2022, background): “Another line of work has focused on training language models to perform reasoning by fine-tuning on gold reasoning traces (Bostrom et al., 2022; Creswell and Shanahan, 2022; Dalvi et al., 2021; Tafjord et al., 2021).”
Language Models can Solve Computer Tasks (2023, background): “Furthermore, various reasoning architectures have been proposed, expanding from naive prompting, such as Selection-Inference [9], Least-to-Most [65], and Faithful reasoning [8].”
Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Prompting (2023, background): “Decomposition-based approaches (Min et al., 2019; Perez et al., 2020; Chen et al., 2022; Creswell and Shanahan, 2022; Tafjord et al., 2022; Eisenstein et al., 2022; Reppert et al., 2023) improve faithfulness by limiting contextual cues that may bias CoT reasoning, with Radhakrishnan et al. (2023)…”
Reasoning on Graphs: Faithful and Interpretable Large Language Model Reasoning (2023, methodology): “To explore faithful reasoning, Entailer (Tafjord et al., 2022) introduces a verifier to validate the reasoning steps generated by LLMs. Creswell & Shanahan (2022) present a framework including two LLMs that are used for selecting and generating reasoning steps, respectively.”
LAMBADA: Backward Chaining for Automated Reasoning in Natural Language (2022, background): “, 2022), sometimes requiring specialized modules trained on intermediate labels (Creswell and Shanahan, 2022).”
Generate rather than Retrieve: Large Language Models are Strong Context Generators (2022, background): “Consideration in combination with recent approaches (Creswell & Shanahan, 2022) to boost generative faithfulness is a also direction worthy of future research.”
Co-Writing Screenplays and Theatre Scripts with Language Models: Evaluation by Industry Professionals (2022, methodology): “Our work of decomposing a log line into a synopsis can be seen as a narrative equivalent of Chain of Thought prompting for reasoning tasks [118], and uses the idea of using LLM output as prompt for the next stage of the generation—also called prompt chaining [121] or language engineering [24].”
Faith and Fate: Limits of Transformers on Compositionality (2023, methodology): “[15] Antonia Creswell and Murray Shanahan.”
Measuring Faithfulness in Chain-of-Thought Reasoning (2023, background): “Creswell & Shanahan (2022) and Creswell et al. (2023) use a language model to choose statements from a context and then make inferences from those selected statements in a separate context window.”