Search

Recent work finds that retrieval-augmented generation with large language models is prone to be influenced by the order of retrieved documents in the context. However, the lack of in-depth analysis limits the use of this phenomenon for prompt engineering in practice. In this study, we posit that likelihoods serve as an effective gauge for language model performance. Through experiments on two question-answering datasets with a variety of state-of-the-art language models, we reveal correlations between answer accuracy and the likelihood of the question at both the corpus level and the instance level. In addition, we find that question likelihood can also indicate the position of the task-relevant information in the context. Based on these findings, we propose two methods that use question likelihood as a gauge for selecting and constructing prompts that lead to better performance. We demonstrate their effectiveness with experiments. In addition, our likelihood-based methods are efficient, as they only need to compute the likelihood of the input, requiring much fewer language model passes than heuristic prompt engineering methods that require generating responses. Our analysis deepens our understanding of how input prompts affect model performance and provides a promising direction for efficient prompt optimization., Comment: Under review at NAACL 2025. Code is available at https://github.com/lyutyuh/poptimizer

Large Language Models (LLMs) can transfer their reasoning skills to smaller models by teaching them to generate the intermediate reasoning process required to solve multistep reasoning tasks. While LLMs can accurately solve reasoning tasks through a variety of strategies, even without fine-tuning, smaller models are not expressive enough to fit the LLMs distribution on all strategies when distilled and tend to prioritize one strategy over the others. This reliance on one strategy poses a challenge for smaller models when attempting to solve reasoning tasks that may be difficult with their preferred strategy. To address this, we propose a distillation method SIKeD (Self-guided Iterative Knowledge Distillation for mathematical reasoning), where the LLM teaches the smaller model to approach a task using different strategies and the smaller model uses its self-generated on-policy outputs to choose the most suitable strategy for the given task. The training continues in a self-guided iterative manner, where for each training iteration, a decision is made on how to combine the LLM data with the self-generated outputs. Unlike traditional distillation methods, SIKeD allows the smaller model to learn which strategy is suitable for a given task while continuously learning to solve a task using different strategies. Our experiments on various mathematical reasoning datasets show that SIKeD significantly outperforms traditional distillation techniques across smaller models of different sizes. Our code is available at: https://github.com/kumar-shridhar/SIKeD

Tasks requiring deductive reasoning, especially those involving multiple steps, often demand adaptive strategies such as intermediate generation of rationales or programs, as no single approach is universally optimal. While Language Models (LMs) can enhance their outputs through iterative self-refinement and strategy adjustments, they frequently fail to apply the most effective strategy in their first attempt. This inefficiency raises the question: Can LMs learn to select the optimal strategy in the first attempt, without a need for refinement? To address this challenge, we introduce SMART (Self-learning Meta-strategy Agent for Reasoning Tasks), a novel framework that enables LMs to autonomously learn and select the most effective strategies for various reasoning tasks. We model the strategy selection process as a Markov Decision Process and leverage reinforcement learning-driven continuous self-improvement to allow the model to find the suitable strategy to solve a given task. Unlike traditional self-refinement methods that rely on multiple inference passes or external feedback, SMART allows an LM to internalize the outcomes of its own reasoning processes and adjust its strategy accordingly, aiming for correct solutions on the first attempt. Our experiments across various reasoning datasets and with different model architectures demonstrate that SMART significantly enhances the ability of models to choose optimal strategies without external guidance (+15 points on the GSM8K dataset). By achieving higher accuracy with a single inference pass, SMART not only improves performance but also reduces computational costs for refinement-based strategies, paving the way for more efficient and intelligent reasoning in LMs.

One strength of modern language models is their ability to incorporate information from a user-input context when answering queries. However, they are not equally sensitive to the subtle changes to that context. To quantify this, Du et al. (2024) gives an information-theoretic metric to measure such sensitivity. Their metric, susceptibility, is defined as the degree to which contexts can influence a model's response to a query at a distributional level. However, exactly computing susceptibility is difficult and, thus, Du et al. (2024) falls back on a Monte Carlo approximation. Due to the large number of samples required, the Monte Carlo approximation is inefficient in practice. As a faster alternative, we propose Fisher susceptibility, an efficient method to estimate the susceptibility based on Fisher information. Empirically, we validate that Fisher susceptibility is comparable to Monte Carlo estimated susceptibility across a diverse set of query domains despite its being $70\times$ faster. Exploiting the improved efficiency, we apply Fisher susceptibility to analyze factors affecting the susceptibility of language models. We observe that larger models are as susceptible as smaller ones.

Large language models (LLMs) can solve arithmetic word problems with high accuracy, but little is known about how well they generalize to problems that are more complex than the ones on which they have been trained. Empirical investigations of such questions are impeded by two major flaws of current evaluations: (i) much of the evaluation data is contaminated, in the sense that it has already been seen during training, and (ii) benchmark datasets do not capture how problem proofs may be arbitrarily complex in various ways. As a step towards addressing these issues, we present a framework for evaluating LLMs on problems with arbitrarily complex arithmetic proofs, called MathGAP. MathGAP generates problems that follow fixed proof specifications -- along with chain-of-thought reasoning annotations -- enabling systematic studies on generalization with respect to arithmetic proof complexity. We apply MathGAP to analyze how in-context learning interacts with generalization to problems that have more complex proofs. We find that among the models tested, most show a significant decrease in performance as proofs get deeper and wider. This effect is more pronounced in complex, nonlinear proof structures, which are challenging even for GPT-4o. Surprisingly, providing in-context examples from the same distribution as the test set is not always beneficial for performance. In particular, zero-shot prompting as well as demonstrating a diverse range of examples that are less complex than the test data sometimes yield similar or higher accuracies., Comment: Preprint

Accurately modeling student cognition is crucial for developing effective AI-driven educational technologies. A key challenge is creating realistic student models that satisfy two essential properties: (1) accurately replicating specific misconceptions, and (2) correctly solving problems where these misconceptions are not applicable. This dual requirement reflects the complex nature of student understanding, where misconceptions coexist with correct knowledge. This paper investigates whether Large Language Models (LLMs) can be instruction-tuned to meet this dual requirement and effectively simulate student thinking in algebra. We introduce MalAlgoPy, a novel Python library that generates datasets reflecting authentic student solution patterns through a graph-based representation of algebraic problem-solving. Utilizing MalAlgoPy, we define and examine Cognitive Student Models (CSMs) - LLMs instruction tuned to faithfully emulate realistic student behavior. Our findings reveal that LLMs trained on misconception examples can efficiently learn to replicate errors. However, the training diminishes the model's ability to solve problems correctly, particularly for problem types where the misconceptions are not applicable, thus failing to satisfy second property of CSMs. We demonstrate that by carefully calibrating the ratio of correct to misconception examples in the training data - sometimes as low as 0.25 - it is possible to develop CSMs that satisfy both properties. Our insights enhance our understanding of AI-based student models and pave the way for effective adaptive learning systems.

One-to-one tutoring is one of the most efficient methods of teaching. Following the rise in popularity of Large Language Models (LLMs), there have been efforts to use them to create conversational tutoring systems, which can make the benefits of one-to-one tutoring accessible to everyone. However, current LLMs are primarily trained to be helpful assistants and thus lack crucial pedagogical skills. For example, they often quickly reveal the solution to the student and fail to plan for a richer multi-turn pedagogical interaction. To use LLMs in pedagogical scenarios, they need to be steered towards using effective teaching strategies: a problem we introduce as Pedagogical Steering and believe to be crucial for the efficient use of LLMs as tutors. We address this problem by formalizing a concept of tutoring strategy, and introducing StratL, an algorithm to model a strategy and use prompting to steer the LLM to follow this strategy. As a case study, we create a prototype tutor for high school math following Productive Failure (PF), an advanced and effective learning design. To validate our approach in a real-world setting, we run a field study with 17 high school students in Singapore. We quantitatively show that StratL succeeds in steering the LLM to follow a Productive Failure tutoring strategy. We also thoroughly investigate the existence of spillover effects on desirable properties of the LLM, like its ability to generate human-like answers. Based on these results, we highlight the challenges in Pedagogical Steering and suggest opportunities for further improvements. We further encourage follow-up research by releasing a dataset of Productive Failure problems and the code of our prototype and algorithm., Comment: 18 pages, 9 figures, 6 tables

Knowledge tracing (KT) is a popular approach for modeling students' learning progress over time, which can enable more personalized and adaptive learning. However, existing KT approaches face two major limitations: (1) they rely heavily on expert-defined knowledge concepts (KCs) in questions, which is time-consuming and prone to errors; and (2) KT methods tend to overlook the semantics of both questions and the given KCs. In this work, we address these challenges and present KCQRL, a framework for automated knowledge concept annotation and question representation learning that can improve the effectiveness of any existing KT model. First, we propose an automated KC annotation process using large language models (LLMs), which generates question solutions and then annotates KCs in each solution step of the questions. Second, we introduce a contrastive learning approach to generate semantically rich embeddings for questions and solution steps, aligning them with their associated KCs via a tailored false negative elimination approach. These embeddings can be readily integrated into existing KT models, replacing their randomly initialized embeddings. We demonstrate the effectiveness of KCQRL across 15 KT algorithms on two large real-world Math learning datasets, where we achieve consistent performance improvements.

Visual language is a system of communication that conveys information through symbols, shapes, and spatial arrangements. Diagrams are a typical example of a visual language depicting complex concepts and their relationships in the form of an image. The symbolic nature of diagrams presents significant challenges for building models capable of understanding them. Yet, recent studies seem to suggest that Large Vision-Language Models (LVLMs) can even tackle complex reasoning tasks involving diagrams. In this paper, we investigate this phenomenon by developing a comprehensive test suite to evaluate the diagram comprehension capability of LVLMs. Our test suite uses a variety of questions focused on concept entities and their relationships over a set of synthetic as well as real diagrams across several domains to evaluate the recognition and reasoning abilities of models. Our evaluation of three LVLMs (GPT-4V, GPT-4o, and Gemini) shows that while these models can accurately identify and reason about entities, their ability to understand relationships is notably limited. Further testing reveals that the decent performance on diagram understanding largely stems from leveraging their background knowledge as shortcuts to identify and reason about the relational information. Thus, we conclude that LVLMs have a limited capability for genuine diagram understanding, and their impressive performance in diagram reasoning is an illusion emanating from other confounding factors, such as the background knowledge in the models.

This work contributes to the scarce empirical literature on LLM-based interactive homework in real-world educational settings and offers a practical, scalable solution for improving homework in schools. Homework is an important part of education in schools across the world, but in order to maximize benefit, it needs to be accompanied with feedback and followup questions. We developed a prompting strategy that enables GPT-4 to conduct interactive homework sessions for high-school students learning English as a second language. Our strategy requires minimal efforts in content preparation, one of the key challenges of alternatives like home tutors or ITSs. We carried out a Randomized Controlled Trial (RCT) in four high-school classes, replacing traditional homework with GPT-4 homework sessions for the treatment group. We observed significant improvements in learning outcomes, specifically a greater gain in grammar, and student engagement. In addition, students reported high levels of satisfaction with the system and wanted to continue using it after the end of the RCT., Comment: Submitted to LAK25

The retrieval augmented generation (RAG) system such as Retro has been shown to improve language modeling capabilities and reduce toxicity and hallucinations by retrieving from a database of non-parametric memory containing trillions of entries. We introduce Retro-li that shows retrieval can also help using a small-scale database, but it demands more accurate and better neighbors when searching in a smaller hence sparser non-parametric memory. This can be met by using a proper semantic similarity search. We further propose adding a regularization to the non-parametric memory for the first time: it significantly reduces perplexity when the neighbor search operations are noisy during inference, and it improves generalization when a domain shift occurs. We also show that Retro-li's non-parametric memory can potentially be implemented on analog in-memory computing hardware, exhibiting O(1) search time while causing noise in retrieving neighbors, with minimal (<1%) performance loss. Our code is available at: https://github.com/IBM/Retrieval-Enhanced-Transformer-Little., Comment: Published as a conference paper at European Conference on Artificial Intelligence 2024

Effective and timely feedback in educational assessments is essential but labor-intensive, especially for complex tasks. Recent developments in automated feedback systems, ranging from deterministic response grading to the evaluation of semi-open and open-ended essays, have been facilitated by advances in machine learning. The emergence of pre-trained Large Language Models, such as GPT-4, offers promising new opportunities for efficiently processing diverse response types with minimal customization. This study evaluates the effectiveness of a pre-trained GPT-4 model in grading semi-open handwritten responses in a university-level mathematics exam. Our findings indicate that GPT-4 provides surprisingly reliable and cost-effective initial grading, subject to subsequent human verification. Future research should focus on refining grading rules and enhancing the extraction of handwritten responses to further leverage these technologies., Comment: 17 pages, 12 figures, 4 tables

Recent advances in large language models (LLMs) have enabled autonomous agents with complex reasoning and task-fulfillment capabilities using a wide range of tools. However, effectively identifying the most relevant tools for a given task becomes a key bottleneck as the toolset size grows, hindering reliable tool utilization. To address this, we introduce Re-Invoke, an unsupervised tool retrieval method designed to scale effectively to large toolsets without training. Specifically, we first generate a diverse set of synthetic queries that comprehensively cover different aspects of the query space associated with each tool document during the tool indexing phase. Second, we leverage LLM's query understanding capabilities to extract key tool-related context and underlying intents from user queries during the inference phase. Finally, we employ a novel multi-view similarity ranking strategy based on intents to pinpoint the most relevant tools for each query. Our evaluation demonstrates that Re-Invoke significantly outperforms state-of-the-art alternatives in both single-tool and multi-tool scenarios, all within a fully unsupervised setting. Notably, on the ToolE datasets, we achieve a 20% relative improvement in nDCG@5 for single-tool retrieval and a 39% improvement for multi-tool retrieval., Comment: EMNLP Findings 2024

We present ALT (ALignment with Textual feedback), an approach that aligns language models with user preferences expressed in text. We argue that text offers greater expressiveness, enabling users to provide richer feedback than simple comparative preferences and this richer feedback can lead to more efficient and effective alignment. ALT aligns the model by conditioning its generation on the textual feedback. Our method relies solely on language modeling techniques and requires minimal hyper-parameter tuning, though it still presents the main benefits of RL-based alignment algorithms and can effectively learn from textual feedback. We explore the efficacy and efficiency of textual feedback across different tasks such as toxicity reduction, summarization, and dialog response generation. We find that ALT outperforms PPO for the task of toxicity reduction while being able to match its performance on summarization with only 20% of the samples. We also explore how ALT can be used with feedback provided by an existing LLM where we explore an LLM providing constrained and unconstrained textual feedback. We also outline future directions to align models with natural language feedback.

Using questions in written text is an effective strategy to enhance readability. However, what makes an active reading question good, what the linguistic role of these questions is, and what is their impact on human reading remains understudied. We introduce GuidingQ, a dataset of 10K in-text questions from textbooks and scientific articles. By analyzing the dataset, we present a comprehensive understanding of the use, distribution, and linguistic characteristics of these questions. Then, we explore various approaches to generate such questions using language models. Our results highlight the importance of capturing inter-question relationships and the challenge of question position identification in generating these questions. Finally, we conduct a human study to understand the implication of such questions on reading comprehension. We find that the generated questions are of high quality and are almost as effective as human-written questions in terms of improving readers' memorization and comprehension., Comment: ACL 2024

Large language models (LLMs) present an opportunity to scale high-quality personalized education to all. A promising approach towards this means is to build dialog tutoring models that scaffold students' problem-solving. However, even though existing LLMs perform well in solving reasoning questions, they struggle to precisely detect student's errors and tailor their feedback to these errors. Inspired by real-world teaching practice where teachers identify student errors and customize their response based on them, we focus on verifying student solutions and show how grounding to such verification improves the overall quality of tutor response generation. We collect a dataset of 1K stepwise math reasoning chains with the first error step annotated by teachers. We show empirically that finding the mistake in a student solution is challenging for current models. We propose and evaluate several verifiers for detecting these errors. Using both automatic and human evaluation we show that the student solution verifiers steer the generation model towards highly targeted responses to student errors which are more often correct with less hallucinations compared to existing baselines., Comment: Preprint. Nico Daheim and Jakub Macina contributed equally. Code and dataset can be found under: https://github.com/eth-lre/verify-then-generate

We evaluate the moral alignment of large language models (LLMs) with human preferences in multilingual trolley problems. Building on the Moral Machine experiment, which captures over 40 million human judgments across 200+ countries, we develop a cross-lingual corpus of moral dilemma vignettes in over 100 languages called MultiTP. This dataset enables the assessment of LLMs' decision-making processes in diverse linguistic contexts. Our analysis explores the alignment of 19 different LLMs with human judgments, capturing preferences across six moral dimensions: species, gender, fitness, status, age, and the number of lives involved. By correlating these preferences with the demographic distribution of language speakers and examining the consistency of LLM responses to various prompt paraphrasings, our findings provide insights into cross-lingual and ethical biases of LLMs and their intersection. We discover significant variance in alignment across languages, challenging the assumption of uniform moral reasoning in AI systems and highlighting the importance of incorporating diverse perspectives in AI ethics. The results underscore the need for further research on the integration of multilingual dimensions in responsible AI research to ensure fair and equitable AI interactions worldwide. Our code and data are at https://github.com/causalNLP/moralmachine

Despite their widespread use, the mechanisms by which large language models (LLMs) represent and regulate uncertainty in next-token predictions remain largely unexplored. This study investigates two critical components believed to influence this uncertainty: the recently discovered entropy neurons and a new set of components that we term token frequency neurons. Entropy neurons are characterized by an unusually high weight norm and influence the final layer normalization (LayerNorm) scale to effectively scale down the logits. Our work shows that entropy neurons operate by writing onto an unembedding null space, allowing them to impact the residual stream norm with minimal direct effect on the logits themselves. We observe the presence of entropy neurons across a range of models, up to 7 billion parameters. On the other hand, token frequency neurons, which we discover and describe here for the first time, boost or suppress each token's logit proportionally to its log frequency, thereby shifting the output distribution towards or away from the unigram distribution. Finally, we present a detailed case study where entropy neurons actively manage confidence in the setting of induction, i.e. detecting and continuing repeated subsequences., Comment: NeurIPS 2024

Retrieval Augmented Generation (RAG) is widely employed to ground responses to queries on domain-specific documents. But do RAG implementations leave out important information when answering queries that need an integrated analysis of information (e.g., Tell me good news in the stock market today.)? To address these concerns, RAG developers need to annotate information retrieval (IR) data for their domain of interest, which is challenging because (1) domain-specific queries usually need nuanced definitions of relevance beyond shallow semantic relevance; and (2) human or GPT-4 annotation is costly and cannot cover all (query, document) pairs (i.e., annotation selection bias), thus harming the effectiveness in evaluating IR recall. To address these challenges, we propose DIRAS (Domain-specific Information Retrieval Annotation with Scalability), a manual-annotation-free schema that fine-tunes open-sourced LLMs to consider nuanced relevance definition and annotate (partial) relevance labels with calibrated relevance scores. Extensive evaluation shows that DIRAS enables smaller (8B) LLMs to achieve GPT-4-level performance on annotating and ranking unseen (query, document) pairs, and is helpful for real-world RAG development. All code, LLM generations, and human annotations can be found in \url{https://github.com/EdisonNi-hku/DIRAS}.

Annually, research teams spend large amounts of money to evaluate the quality of machine translation systems (WMT, inter alia). This is expensive because it requires a lot of expert human labor. The recently adopted annotation protocol, Error Span Annotation (ESA), has annotators marking erroneous parts of the translation and then assigning a final score. A lot of the annotator time is spent on scanning the translation for possible errors. In our work, we help the annotators by pre-filling the error annotations with recall-oriented automatic quality estimation. With this AI assistance, we obtain annotations at the same quality level while cutting down the time per span annotation by half (71s/error span $\rightarrow$ 31s/error span). The biggest advantage of ESA$^\mathrm{AI}$ protocol is an accurate priming of annotators (pre-filled error spans) before they assign the final score. This also alleviates a potential automation bias, which we confirm to be low. In addition, the annotation budget can be reduced by almost 25\% with filtering of examples that the AI deems to be very likely to be correct.

Long-context language models (LCLMs) have the potential to revolutionize our approach to tasks traditionally reliant on external tools like retrieval systems or databases. Leveraging LCLMs' ability to natively ingest and process entire corpora of information offers numerous advantages. It enhances user-friendliness by eliminating the need for specialized knowledge of tools, provides robust end-to-end modeling that minimizes cascading errors in complex pipelines, and allows for the application of sophisticated prompting techniques across the entire system. To assess this paradigm shift, we introduce LOFT, a benchmark of real-world tasks requiring context up to millions of tokens designed to evaluate LCLMs' performance on in-context retrieval and reasoning. Our findings reveal LCLMs' surprising ability to rival state-of-the-art retrieval and RAG systems, despite never having been explicitly trained for these tasks. However, LCLMs still face challenges in areas like compositional reasoning that are required in SQL-like tasks. Notably, prompting strategies significantly influence performance, emphasizing the need for continued research as context lengths grow. Overall, LOFT provides a rigorous testing ground for LCLMs, showcasing their potential to supplant existing paradigms and tackle novel tasks as model capabilities scale., Comment: 29 pages. Dataset available at https://github.com/google-deepmind/loft

High-quality Machine Translation (MT) evaluation relies heavily on human judgments. Comprehensive error classification methods, such as Multidimensional Quality Metrics (MQM), are expensive as they are time-consuming and can only be done by experts, whose availability may be limited especially for low-resource languages. On the other hand, just assigning overall scores, like Direct Assessment (DA), is simpler and faster and can be done by translators of any level, but is less reliable. In this paper, we introduce Error Span Annotation (ESA), a human evaluation protocol which combines the continuous rating of DA with the high-level error severity span marking of MQM. We validate ESA by comparing it to MQM and DA for 12 MT systems and one human reference translation (English to German) from WMT23. The results show that ESA offers faster and cheaper annotations than MQM at the same quality level, without the requirement of expensive MQM experts.

Large language models (LLMs) exhibit an intriguing ability to learn a novel task from in-context examples presented in a demonstration, termed in-context learning (ICL). Understandably, a swath of research has been dedicated to uncovering the theories underpinning ICL. One popular hypothesis explains ICL by task selection. LLMs identify the task based on the demonstration and generalize it to the prompt. Another popular hypothesis is that ICL is a form of meta-learning, i.e., the models learn a learning algorithm at pre-training time and apply it to the demonstration. Finally, a third hypothesis argues that LLMs use the demonstration to select a composition of tasks learned during pre-training to perform ICL. In this paper, we empirically explore these three hypotheses that explain LLMs' ability to learn in context with a suite of experiments derived from common text classification tasks. We invalidate the first two hypotheses with counterexamples and provide evidence in support of the last hypothesis. Our results suggest an LLM could learn a novel task in context via composing tasks learned during pre-training., Comment: This work is published in ACL 2024

Pre-trained language encoders -- functions that represent text as vectors -- are an integral component of many NLP tasks. We tackle a natural question in language encoder analysis: What does it mean for two encoders to be similar? We contend that a faithful measure of similarity needs to be \emph{intrinsic}, that is, task-independent, yet still be informative of \emph{extrinsic} similarity -- the performance on downstream tasks. It is common to consider two encoders similar if they are \emph{homotopic}, i.e., if they can be aligned through some transformation. In this spirit, we study the properties of \emph{affine} alignment of language encoders and its implications on extrinsic similarity. We find that while affine alignment is fundamentally an asymmetric notion of similarity, it is still informative of extrinsic similarity. We confirm this on datasets of natural language representations. Beyond providing useful bounds on extrinsic similarity, affine intrinsic similarity also allows us to begin uncovering the structure of the space of pre-trained encoders by defining an order over them., Comment: 10 pages

The recent development of Large Language Models (LLMs) has changed our role in interacting with them. Instead of primarily testing these models with questions we already know the answers to, we now use them to explore questions where the answers are unknown to us. This shift, which hasn't been fully addressed in existing datasets, highlights the growing need to understand naturally occurring human questions - that are more complex, open-ended, and reflective of real-world needs. To this end, we present NatQuest, a collection of 13,500 naturally occurring questions from three diverse sources: human-to-search-engine queries, human-to-human interactions, and human-to-LLM conversations. Our comprehensive collection enables a rich understanding of human curiosity across various domains and contexts. Our analysis reveals a significant presence of causal questions (up to 42%) within the dataset, for which we develop an iterative prompt improvement framework to identify all causal queries, and examine their unique linguistic properties, cognitive complexity, and source distribution. We also lay the groundwork to explore LLM performance on these questions and provide six efficient classification models to identify causal questions at scale for future work.

Implicit Personalization (IP) is a phenomenon of language models inferring a user's background from the implicit cues in the input prompts and tailoring the response based on this inference. While previous work has touched upon various instances of this problem, there lacks a unified framework to study this behavior. This work systematically studies IP through a rigorous mathematical formulation, a multi-perspective moral reasoning framework, and a set of case studies. Our theoretical foundation for IP relies on a structural causal model and introduces a novel method, indirect intervention, to estimate the causal effect of a mediator variable that cannot be directly intervened upon. Beyond the technical approach, we also introduce a set of moral reasoning principles based on three schools of moral philosophy to study when IP may or may not be ethically appropriate. Equipped with both mathematical and ethical insights, we present three diverse case studies illustrating the varied nature of the IP problem and offer recommendations for future research. Our code is at https://github.com/jiarui-liu/IP, and our data is at https://huggingface.co/datasets/Jerry999/ImplicitPersonalizationData., Comment: EMNLP 2024 Findings

In this paper, we investigate the Sombor index of the total graph and unit graph of $\mathbb{Z}_n$ which is denoted by $T_{\Gamma}(\mathbb{Z}_n)$ and $G(\mathbb{Z}_n)$ respectively for $n \in \{2k, p^{\alpha}, pq, p^2q\}$ where $p$ and $q$ are distinct odd prime numbers such that $p < q$. Moreover, we compute the Sombor index of any finite local ring., Comment: 11 pages

Natural languages are believed to be (mildly) context-sensitive. Despite underpinning remarkably capable large language models, transformers are unable to model many context-free language tasks. In an attempt to address this limitation in the modeling power of transformer-based language models, we propose augmenting them with a differentiable, stack-based attention mechanism. Our stack-based attention mechanism can be incorporated into any transformer-based language model and adds a level of interpretability to the model. We show that the addition of our stack-based attention mechanism enables the transformer to model some, but not all, deterministic context-free languages., Comment: NAACL 2024 Findings

As AI systems pervade human life, ensuring that large language models (LLMs) make safe decisions remains a significant challenge. We introduce the Governance of the Commons Simulation (GovSim), a generative simulation platform designed to study strategic interactions and cooperative decision-making in LLMs. In GovSim, a society of AI agents must collectively balance exploiting a common resource with sustaining it for future use. This environment enables the study of how ethical considerations, strategic planning, and negotiation skills impact cooperative outcomes. We develop an LLM-based agent architecture and test it with the leading open and closed LLMs. We find that all but the most powerful LLM agents fail to achieve a sustainable equilibrium in GovSim, with the highest survival rate below 54%. Ablations reveal that successful multi-agent communication between agents is critical for achieving cooperation in these cases. Furthermore, our analyses show that the failure to achieve sustainable cooperation in most LLMs stems from their inability to formulate and analyze hypotheses about the long-term effects of their actions on the equilibrium of the group. Finally, we show that agents that leverage "Universalization"-based reasoning, a theory of moral thinking, are able to achieve significantly better sustainability. Taken together, GovSim enables us to study the mechanisms that underlie sustainable self-government with specificity and scale. We open source the full suite of our research results, including the simulation environment, agent prompts, and a comprehensive web interface., Comment: Revised version

Logical fallacies are common errors in reasoning that undermine the logic of an argument. Automatically detecting logical fallacies has important applications in tracking misinformation and validating claims. In this paper, we design a process to reliably detect logical fallacies by translating natural language to First-order Logic (FOL) step-by-step using Large Language Models (LLMs). We then utilize Satisfiability Modulo Theory (SMT) solvers to reason about the validity of the formula and classify inputs as either a fallacy or valid statement. Our model also provides a novel means of utilizing LLMs to interpret the output of the SMT solver, offering insights into the counter-examples that illustrate why a given sentence is considered a logical fallacy. Our approach is robust, interpretable and does not require training data or fine-tuning. We evaluate our model on a mixed dataset of fallacies and valid sentences. The results demonstrate improved performance compared to end-to-end LLMs, with our classifier achieving an F1-score of 71\% on the Logic dataset. The approach is able to generalize effectively, achieving an F1-score of 73% on the challenge set, LogicClimate, outperforming state-of-the-art models by 21% despite its much smaller size.

Sentiment analysis (SA) aims to identify the sentiment expressed in a text, such as a product review. Given a review and the sentiment associated with it, this work formulates SA as a combination of two tasks: (1) a causal discovery task that distinguishes whether a review "primes" the sentiment (Causal Hypothesis C1), or the sentiment "primes" the review (Causal Hypothesis C2); and (2) the traditional prediction task to model the sentiment using the review as input. Using the peak-end rule in psychology, we classify a sample as C1 if its overall sentiment score approximates an average of all the sentence-level sentiments in the review, and C2 if the overall sentiment score approximates an average of the peak and end sentiments. For the prediction task, we use the discovered causal mechanisms behind the samples to improve LLM performance by proposing causal prompts that give the models an inductive bias of the underlying causal graph, leading to substantial improvements by up to 32.13 F1 points on zero-shot five-class SA. Our code is at https://github.com/cogito233/causal-sa, Comment: EMNLP 2024 Findings

In this report, we introduce the Gemini 1.5 family of models, representing the next generation of highly compute-efficient multimodal models capable of recalling and reasoning over fine-grained information from millions of tokens of context, including multiple long documents and hours of video and audio. The family includes two new models: (1) an updated Gemini 1.5 Pro, which exceeds the February version on the great majority of capabilities and benchmarks; (2) Gemini 1.5 Flash, a more lightweight variant designed for efficiency with minimal regression in quality. Gemini 1.5 models achieve near-perfect recall on long-context retrieval tasks across modalities, improve the state-of-the-art in long-document QA, long-video QA and long-context ASR, and match or surpass Gemini 1.0 Ultra's state-of-the-art performance across a broad set of benchmarks. Studying the limits of Gemini 1.5's long-context ability, we find continued improvement in next-token prediction and near-perfect retrieval (>99%) up to at least 10M tokens, a generational leap over existing models such as Claude 3.0 (200k) and GPT-4 Turbo (128k). Finally, we highlight real-world use cases, such as Gemini 1.5 collaborating with professionals on completing their tasks achieving 26 to 75% time savings across 10 different job categories, as well as surprising new capabilities of large language models at the frontier; when given a grammar manual for Kalamang, a language with fewer than 200 speakers worldwide, the model learns to translate English to Kalamang at a similar level to a person who learned from the same content.

Educational chatbots are a promising tool for assisting student learning. However, the development of effective chatbots in education has been challenging, as high-quality data is seldom available in this domain. In this paper, we propose a framework for generating synthetic teacher-student interactions grounded in a set of textbooks. Our approaches capture one aspect of learning interactions where curious students with partial knowledge interactively ask a teacher questions about the material in the textbook. We highlight various quality criteria that such dialogues should fulfill and compare several approaches relying on either prompting or fine-tuning large language models. We use synthetic dialogues to train educational chatbots and show benefits of further fine-tuning in different educational domains. However, human evaluation shows that our best data synthesis method still suffers from hallucinations and tends to reiterate information from previous conversations. Our findings offer insights for future efforts in synthesizing conversational data that strikes a balance between size and quality. We will open-source our data and code., Comment: 24 pages, 19 tables, 2 figures

Accurately gauging the confidence level of Large Language Models' (LLMs) predictions is pivotal for their reliable application. However, LLMs are often uncalibrated inherently and elude conventional calibration techniques due to their proprietary nature and massive scale. In this work, we explore the potential of deriving confidence from the distribution of multiple randomly sampled model generations, via three measures of consistency. We perform an extensive evaluation across various open and closed-source models on nine reasoning datasets. Results show that consistency-based calibration methods outperform existing post-hoc approaches. Meanwhile, we find that factors such as intermediate explanations, model scaling, and larger sample sizes enhance calibration, while instruction-tuning makes calibration more difficult. Moreover, confidence scores obtained from consistency have the potential to enhance model performance. Finally, we offer practical guidance on choosing suitable consistency metrics for calibration, tailored to the characteristics of various LMs.

Interpretability research aims to bridge the gap between empirical success and our scientific understanding of the inner workings of large language models (LLMs). However, most existing research focuses on analyzing a single mechanism, such as how models copy or recall factual knowledge. In this work, we propose a formulation of competition of mechanisms, which focuses on the interplay of multiple mechanisms instead of individual mechanisms and traces how one of them becomes dominant in the final prediction. We uncover how and where mechanisms compete within LLMs using two interpretability methods: logit inspection and attention modification. Our findings show traces of the mechanisms and their competition across various model components and reveal attention positions that effectively control the strength of certain mechanisms. Code: https://github.com/francescortu/comp-mech. Data: https://huggingface.co/datasets/francescortu/comp-mech., Comment: ACL 2024

With the rise of generative AI, automated fact-checking methods to combat misinformation are becoming more and more important. However, factual claim detection, the first step in a fact-checking pipeline, suffers from two key issues that limit its scalability and generalizability: (1) inconsistency in definitions of the task and what a claim is, and (2) the high cost of manual annotation. To address (1), we review the definitions in related work and propose a unifying definition of factual claims that focuses on verifiability. To address (2), we introduce AFaCTA (Automatic Factual Claim deTection Annotator), a novel framework that assists in the annotation of factual claims with the help of large language models (LLMs). AFaCTA calibrates its annotation confidence with consistency along three predefined reasoning paths. Extensive evaluation and experiments in the domain of political speech reveal that AFaCTA can efficiently assist experts in annotating factual claims and training high-quality classifiers, and can work with or without expert supervision. Our analyses also result in PoliClaim, a comprehensive claim detection dataset spanning diverse political topics., Comment: ACL2024 Main Conference

Large Language Models (LLMs) have found several use cases in education, ranging from automatic question generation to essay evaluation. In this paper, we explore the potential of using Large Language Models (LLMs) to author Intelligent Tutoring Systems. A common pitfall of LLMs is their straying from desired pedagogical strategies such as leaking the answer to the student, and in general, providing no guarantees. We posit that while LLMs with certain guardrails can take the place of subject experts, the overall pedagogical design still needs to be handcrafted for the best learning results. Based on this principle, we create a sample end-to-end tutoring system named MWPTutor, which uses LLMs to fill in the state space of a pre-defined finite state transducer. This approach retains the structure and the pedagogy of traditional tutoring systems that has been developed over the years by learning scientists but brings in additional flexibility of LLM-based approaches. Through a human evaluation study on two datasets based on math word problems, we show that our hybrid approach achieves a better overall tutoring score than an instructed, but otherwise free-form, GPT-4. MWPTutor is completely modular and opens up the scope for the community to improve its performance by improving individual modules or using different teaching strategies that it can follow., Comment: To be presented at Learning@Scale 2024

Common methods for aligning large language models (LLMs) with desired behaviour heavily rely on human-labelled data. However, as models grow increasingly sophisticated, they will surpass human expertise, and the role of human evaluation will evolve into non-experts overseeing experts. In anticipation of this, we ask: can weaker models assess the correctness of stronger models? We investigate this question in an analogous setting, where stronger models (experts) possess the necessary information to answer questions and weaker models (non-experts) lack this information. The method we evaluate is debate, where two LLM experts each argue for a different answer, and a non-expert selects the answer. We find that debate consistently helps both non-expert models and humans answer questions, achieving 76% and 88% accuracy respectively (naive baselines obtain 48% and 60%). Furthermore, optimising expert debaters for persuasiveness in an unsupervised manner improves non-expert ability to identify the truth in debates. Our results provide encouraging empirical evidence for the viability of aligning models with debate in the absence of ground truth., Comment: For code please check: https://github.com/ucl-dark/llm_debate

There is increasing interest in employing large language models (LLMs) as cognitive models. For such purposes, it is central to understand which properties of human cognition are well-modeled by LLMs, and which are not. In this work, we study the biases of LLMs in relation to those known in children when solving arithmetic word problems. Surveying the learning science literature, we posit that the problem-solving process can be split into three distinct steps: text comprehension, solution planning and solution execution. We construct tests for each one in order to understand whether current LLMs display the same cognitive biases as children in these steps. We generate a novel set of word problems for each of these tests, using a neuro-symbolic approach that enables fine-grained control over the problem features. We find evidence that LLMs, with and without instruction-tuning, exhibit human-like biases in both the text-comprehension and the solution-planning steps of the solving process, but not in the final step, in which the arithmetic expressions are executed to obtain the answer., Comment: Accepted at ICML 2024

Humans are capable of strategically deceptive behavior: behaving helpfully in most situations, but then behaving very differently in order to pursue alternative objectives when given the opportunity. If an AI system learned such a deceptive strategy, could we detect it and remove it using current state-of-the-art safety training techniques? To study this question, we construct proof-of-concept examples of deceptive behavior in large language models (LLMs). For example, we train models that write secure code when the prompt states that the year is 2023, but insert exploitable code when the stated year is 2024. We find that such backdoor behavior can be made persistent, so that it is not removed by standard safety training techniques, including supervised fine-tuning, reinforcement learning, and adversarial training (eliciting unsafe behavior and then training to remove it). The backdoor behavior is most persistent in the largest models and in models trained to produce chain-of-thought reasoning about deceiving the training process, with the persistence remaining even when the chain-of-thought is distilled away. Furthermore, rather than removing backdoors, we find that adversarial training can teach models to better recognize their backdoor triggers, effectively hiding the unsafe behavior. Our results suggest that, once a model exhibits deceptive behavior, standard techniques could fail to remove such deception and create a false impression of safety., Comment: updated to add missing acknowledgements