ReasonAgain: Using Extractable Symbolic Programs to Evaluate Mathematical Reasoning

Existing math datasets evaluate the reasoning abilities of large language models (LLMs) by either using the final answer or the intermediate reasoning steps derived from static examples. However, the former approach fails to surface model's uses of shortcuts and wrong reasoning while the later poses challenges in accommodating alternative solutions. In this work, we seek to use symbolic programs as a means for automated evaluation if a model can consistently produce correct final answers across various inputs to the program. We begin by extracting programs for popular math datasets (GSM8K and MATH) using GPT4-o. For those executable programs verified using the original input-output pairs, they are found to encapsulate the proper reasoning required to solve the original text questions. We then prompt GPT4-o to generate new questions using alternative input-output pairs based the extracted program. We apply the resulting datasets to evaluate a collection of LLMs. In our experiments, we observe significant accuracy drops using our proposed evaluation compared with original static examples, suggesting the fragility of math reasoning in state-of-the-art LLMs.

ToW: Thoughts of Words Improve Reasoning in Large Language Models

We introduce thoughts of words (ToW), a novel training-time data-augmentation method for next-word prediction. ToW views next-word prediction as a core reasoning task and injects fine-grained thoughts explaining what the next word should be and how it is related to the previous contexts in pre-training texts. Our formulation addresses two fundamental drawbacks of existing next-word prediction learning schemes: they induce factual hallucination and are inefficient for models to learn the implicit reasoning processes in raw texts. While there are many ways to acquire such thoughts of words, we explore the first step of acquiring ToW annotations through distilling from larger models. After continual pre-training with only 70K ToW annotations, we effectively improve models' reasoning performances by 7% to 9% on average and reduce model hallucination by up to 10%. At the same time, ToW is entirely agnostic to tasks and applications, introducing no additional biases on labels or semantics.

FamiCom: Further Demystifying Prompts for Language Models with Task-Agnostic Performance Estimation

Language models have shown impressive in-context-learning capabilities, which allow them to benefit from input prompts and perform better on downstream end tasks. Existing works investigate the mechanisms behind this observation, and propose label-agnostic prompt metrics that can better estimate end-task performances. One popular approach is using perplexity as a way to measure models' familiarity with the prompt. While showing consistent improvements on in-domain tasks, we found that familiarity metrics such as perplexity cannot accurately estimate performance in complicated situations such as task or domain transferring scenarios. In this work, we propose a revised measure called FamiCom, providing a more comprehensive measure for task-agnostic performance estimation. Specifically, FamiCom combines familiarity with \textit{complexity} -- the inherent difficulty of end tasks, which is an important factor missing from current metrics. Experiments show that FamiCom strongly correlates with end-task performances, producing a 0.85 Spearman's correlation, versus 0.43 of familiarity-only ones'. We further apply FamiCom to automatic prompt and demonstration selection, and outperform existing methods and baselines by more than 7.0% in accuracy.

BIRD: A Trustworthy Bayesian Inference Framework for Large Language Models

Large language models primarily rely on inductive reasoning for decision making. This results in unreliable decisions when applied to real-world tasks that often present incomplete contexts and conditions. Thus, accurate probability estimation and appropriate interpretations are required to enhance decision-making reliability. In this paper, we propose a Bayesian inference framework called BIRD for large language models. BIRD provides controllable and interpretable probability estimation for model decisions, based on abductive factors, LLM entailment, as well as learnable deductive Bayesian modeling. Experiments show that BIRD produces probability estimations that align with human judgments over 65% of the time using open-sourced Llama models, outperforming the state-of-the-art GPT-4 by 35%. We also show that BIRD can be directly used for trustworthy decision making on many real-world applications.

Conceptual and Unbiased Reasoning in Language Models

Conceptual reasoning, the ability to reason in abstract and high-level perspectives, is key to generalization in human cognition. However, limited study has been done on large language models' capability to perform conceptual reasoning. In this work, we bridge this gap and propose a novel conceptualization framework that forces models to perform conceptual reasoning on abstract questions and generate solutions in a verifiable symbolic space. Using this framework as an analytical tool, we show that existing large language models fall short on conceptual reasoning, dropping 9% to 28% on various benchmarks compared to direct inference methods. We then discuss how models can improve since high-level abstract reasoning is key to unbiased and generalizable decision-making. We propose two techniques to add trustworthy induction signals by generating familiar questions with similar underlying reasoning paths and asking models to perform self-refinement. Experiments show that our proposed techniques improve models' conceptual reasoning performance by 8% to 11%, achieving a more robust reasoning system that relies less on inductive biases.

Cognitive Overload: Jailbreaking Large Language Models with Overloaded Logical Thinking

While large language models (LLMs) have demonstrated increasing power, they have also given rise to a wide range of harmful behaviors. As representatives, jailbreak attacks can provoke harmful or unethical responses from LLMs, even after safety alignment. In this paper, we investigate a novel category of jailbreak attacks specifically designed to target the cognitive structure and processes of LLMs. Specifically, we analyze the safety vulnerability of LLMs in the face of (1) multilingual cognitive overload, (2) veiled expression, and (3) effect-to-cause reasoning. Different from previous jailbreak attacks, our proposed cognitive overload is a black-box attack with no need for knowledge of model architecture or access to model weights. Experiments conducted on AdvBench and MasterKey reveal that various LLMs, including both popular open-source model Llama 2 and the proprietary model ChatGPT, can be compromised through cognitive overload. Motivated by cognitive psychology work on managing cognitive load, we further investigate defending cognitive overload attack from two perspectives. Empirical studies show that our cognitive overload from three perspectives can jailbreak all studied LLMs successfully, while existing defense strategies can hardly mitigate the caused malicious uses effectively.

Deceiving Semantic Shortcuts on Reasoning Chains: How Far Can Models Go without Hallucination?

Despite the recent advancement in large language models (LLMs) and their high performances across numerous benchmarks, recent research has unveiled that LLMs suffer from hallucinations and unfaithful reasoning. This work studies a specific type of hallucination induced by semantic associations. Specifically, we investigate to what extent LLMs take shortcuts from certain keyword/entity biases in the prompt instead of following the correct reasoning path. To quantify this phenomenon, we propose a novel probing method and benchmark called EureQA. We start from questions that LLMs will answer correctly with utmost certainty, and mask the important entity with evidence sentence recursively, asking models to find masked entities according to a chain of evidence before answering the question. During the construction of the evidence, we purposefully replace semantic clues (entities) that may lead to the correct answer with distractor clues (evidence) that will not directly lead to the correct answer but require a chain-like reasoning process. We evaluate if models can follow the correct reasoning chain instead of short-cutting through distractor clues. We find that existing LLMs lack the necessary capabilities to follow correct reasoning paths and resist the attempt of greedy shortcuts. We show that the distractor semantic associations often lead to model hallucination, which is strong evidence that questions the validity of current LLM reasoning.

Sub-Sentence Encoder: Contrastive Learning of Propositional Semantic Representations

We introduce sub-sentence encoder, a contrastively-learned contextual embedding model for fine-grained semantic representation of text. In contrast to the standard practice with sentence embeddings, where the meaning of an entire sequence of text is encoded into a fixed-length vector, the sub-sentence encoder learns to produce distinct contextual embeddings corresponding to different atomic propositions, i.e. atomic units of meaning expressed within a text sequence. The sub-sentence embeddings are contrastively learned to recognize (inferred) semantic equivalence between propositions across different text sequences. Our experiments show the effectiveness of sub-sentence encoders in applications, such as retrieving supporting facts for fine-grained text attribution or recognizing the conditional semantic similarity between texts. In practice, we demonstrate that sub-sentence encoders keep the same level of inference cost and space complexity compared to sentence encoders.

Building Interpretable and Reliable Open Information Retriever for New Domains Overnight

Information retrieval (IR) or knowledge retrieval, is a critical component for many down-stream tasks such as open-domain question answering (QA). It is also very challenging, as it requires succinctness, completeness, and correctness. In recent works, dense retrieval models have achieved state-of-the-art (SOTA) performance on in-domain IR and QA benchmarks by representing queries and knowledge passages with dense vectors and learning the lexical and semantic similarity. However, using single dense vectors and end-to-end supervision are not always optimal because queries may require attention to multiple aspects and event implicit knowledge. In this work, we propose an information retrieval pipeline that uses entity/event linking model and query decomposition model to focus more accurately on different information units of the query. We show that, while being more interpretable and reliable, our proposed pipeline significantly improves passage coverages and denotation accuracies across five IR and QA benchmarks. It will be the go-to system to use for applications that need to perform IR on a new domain without much dedicated effort, because of its superior interpretability and cross-domain performance.

Interpretable by Design Visual Question Answering

Model interpretability has long been a hard problem for the AI community especially in the multimodal setting, where vision and language need to be aligned and reasoned at the same time. In this paper, we specifically focus on the problem of Visual Question Answering (VQA). While previous researches try to probe into the network structures of black-box multimodal models, we propose to tackle the problem from a different angle -- to treat interpretability as an explicit additional goal. Given an image and question, we argue that an interpretable VQA model should be able to tell what conclusions it can get from which part of the image, and show how each statement help to arrive at an answer. We introduce InterVQA: Interpretable-by-design VQA, where we design an explicit intermediate dynamic reasoning structure for VQA problems and enforce symbolic reasoning that only use the structure for final answer prediction to take place. InterVQA produces high-quality explicit intermediate reasoning steps, while maintaining similar to the state-of-the-art (sota) end-task performance.