We propose a method to optimize language model pre-training data mixtures through efficient approximation of the cross-entropy loss corresponding to each candidate mixture via a Mixture of Data Experts (MDE). We use this approximation as a source of additional features in a regression model, trained from observations of model loss for a small number of mixtures. Experiments with Transformer decoder-only language models in the range of 70M to 1B parameters on the SlimPajama dataset show that our method achieves significantly better performance than approaches that train regression models using only the mixture rates as input features. Combining this improved optimization method with an objective that takes into account cross-entropy on end task data leads to superior performance on few-shot downstream evaluations. We also provide theoretical insights on why aggregation of data expert predictions can provide good approximations to model losses for data mixtures.
An increasingly prevalent socio-technical problem is people being taken in by offers that sound ``too good to be true'', where persuasion and trust shape decision-making. This paper investigates how \abr{ai} can help detect these deceptive scenarios. We analyze how humans strategically deceive each other in \textit{Diplomacy}, a board game that requires both natural language communication and strategic reasoning. This requires extracting logical forms of proposed agreements in player communications and computing the relative rewards of the proposal using agents' value functions. Combined with text-based features, this can improve our deception detection. Our method detects human deception with a high precision when compared to a Large Language Model approach that flags many true messages as deceptive. Future human-\abr{ai} interaction tools can build on our methods for deception detection by triggering \textit{friction} to give users a chance of interrogating suspicious proposals.
Retrieval-augmented generation (RAG) has emerged to address the knowledge-intensive visual question answering (VQA) task. Current methods mainly employ separate retrieval and generation modules to acquire external knowledge and generate answers, respectively. We propose ReAuSE, an alternative to the previous RAG model for the knowledge-based VQA task, which seamlessly integrates knowledge retriever into the generative multi-modal large language model, serving as a built-in search engine. Specifically, our model functions both as a generative retriever and an accurate answer generator. It not only helps retrieve documents from the knowledge base by producing identifiers for each document, but it also answers visual questions based on the retrieved documents. Furthermore, we propose a reinforced retrieval calibration module from relevance feedback to improve retrieval performance and align with the preferences for accurate answer generation. Extensive experiments on two representative OKVQA and A-OKVQA datasets demonstrate significant improvements ranging from 2.9\% to 9.6\% across all evaluation metrics when compared to strong baselines.
Reinforcement learning from human feedback (RLHF) has emerged as the primary method for aligning large language models (LLMs) with human preferences. While it enables LLMs to achieve human-level alignment, it often incurs significant computational and financial costs due to its reliance on training external reward models or human-labeled preferences. In this work, we propose \textbf{Implicit Preference Optimization (IPO)}, an alternative approach that leverages generative LLMs as preference classifiers, thereby reducing the dependence on external human feedback or reward models to obtain preferences. We conduct a comprehensive evaluation on the preference classification ability of LLMs using RewardBench, assessing models across different sizes, architectures, and training levels to validate our hypothesis. Furthermore, we investigate the self-improvement capabilities of LLMs by generating multiple responses for a given instruction and employing the model itself as a preference classifier for Direct Preference Optimization (DPO)-based training. Our findings demonstrate that models trained through IPO achieve performance comparable to those utilizing state-of-the-art reward models for obtaining preferences.
In this study, we investigate the integration of a large language model (LLM) with an automatic speech recognition (ASR) system, specifically focusing on enhancing rare word recognition performance. Using a 190,000-hour dataset primarily sourced from YouTube, pre-processed with Whisper V3 pseudo-labeling, we demonstrate that the LLM-ASR architecture outperforms traditional Zipformer-Transducer models in the zero-shot rare word recognition task, after training on a large dataset. Our analysis reveals that the LLM contributes significantly to improvements in rare word error rate (R-WER), while the speech encoder primarily determines overall transcription performance (Orthographic Word Error Rate, O-WER, and Normalized Word Error Rate, N-WER). Through extensive ablation studies, we highlight the importance of adapter integration in aligning speech encoder outputs with the LLM's linguistic capabilities. Furthermore, we emphasize the critical role of high-quality labeled data in achieving optimal performance. These findings provide valuable insights into the synergy between LLM-based ASR architectures, paving the way for future advancements in large-scale LLM-based speech recognition systems.
Large language models have shown remarkable reasoning abilities and scaling laws suggest that large parameter count, especially along the depth axis, is the primary driver. In this work, we make a stronger claim -- many reasoning problems require a large depth but not necessarily many parameters. This unlocks a novel application of looped models for reasoning. Firstly, we show that for many synthetic reasoning problems like addition, $p$-hop induction, and math problems, a $k$-layer transformer looped $L$ times nearly matches the performance of a $kL$-layer non-looped model, and is significantly better than a $k$-layer model. This is further corroborated by theoretical results showing that many such reasoning problems can be solved via iterative algorithms, and thus, can be solved effectively using looped models with nearly optimal depth. Perhaps surprisingly, these benefits also translate to practical settings of language modeling -- on many downstream reasoning tasks, a language model with $k$-layers looped $L$ times can be competitive to, if not better than, a $kL$-layer language model. In fact, our empirical analysis reveals an intriguing phenomenon: looped and non-looped models exhibit scaling behavior that depends on their effective depth, akin to the inference-time scaling of chain-of-thought (CoT) reasoning. We further elucidate the connection to CoT reasoning by proving that looped models implicitly generate latent thoughts and can simulate $T$ steps of CoT with $T$ loops. Inspired by these findings, we also present an interesting dichotomy between reasoning and memorization, and design a looping-based regularization that is effective on both fronts.
How can we verify whether copyrighted content was used to train a large vision-language model (VLM) without direct access to its training data? Motivated by the hypothesis that a VLM is able to recognize images from its training corpus, we propose DIS-CO, a novel approach to infer the inclusion of copyrighted content during the model's development. By repeatedly querying a VLM with specific frames from targeted copyrighted material, DIS-CO extracts the content's identity through free-form text completions. To assess its effectiveness, we introduce MovieTection, a benchmark comprising 14,000 frames paired with detailed captions, drawn from films released both before and after a model's training cutoff. Our results show that DIS-CO significantly improves detection performance, nearly doubling the average AUC of the best prior method on models with logits available. Our findings also highlight a broader concern: all tested models appear to have been exposed to some extent to copyrighted content. Our code and data are available at https://github.com/avduarte333/DIS-CO
Surface contamination on electrical grid insulators leads to an increase in leakage current until an electrical discharge occurs, which can result in a power system shutdown. To mitigate the possibility of disruptive faults resulting in a power outage, monitoring contamination and leakage current can help predict the progression of faults. Given this need, this paper proposes a hybrid deep learning (DL) model for predicting the increase in leakage current in high-voltage insulators. The hybrid structure considers a multi-criteria optimization using tree-structured Parzen estimation, an input stage filter for signal noise attenuation combined with a large language model (LLM) applied for time series forecasting. The proposed optimized LLM outperforms state-of-the-art DL models with a root-mean-square error equal to 2.24$\times10^{-4}$ for a short-term horizon and 1.21$\times10^{-3}$ for a medium-term horizon.
Large language models ($\textbf{LLMs}$) have emerged as a powerful method for discovery. Instead of utilizing numerical data, LLMs utilize associated variable $\textit{semantic metadata}$ to predict variable relationships. Simultaneously, LLMs demonstrate impressive abilities to act as black-box optimizers when given an objective $f$ and sequence of trials. We study LLMs at the intersection of these two capabilities by applying LLMs to the task of $\textit{interactive graph discovery}$: given a ground truth graph $G^*$ capturing variable relationships and a budget of $I$ edge experiments over $R$ rounds, minimize the distance between the predicted graph $\hat{G}_R$ and $G^*$ at the end of the $R$-th round. To solve this task we propose $\textbf{IGDA}$, a LLM-based pipeline incorporating two key components: 1) an LLM uncertainty-driven method for edge experiment selection 2) a local graph update strategy utilizing binary feedback from experiments to improve predictions for unselected neighboring edges. Experiments on eight different real-world graphs show our approach often outperforms all baselines including a state-of-the-art numerical method for interactive graph discovery. Further, we conduct a rigorous series of ablations dissecting the impact of each pipeline component. Finally, to assess the impact of memorization, we apply our interactive graph discovery strategy to a complex, new (as of July 2024) causal graph on protein transcription factors, finding strong performance in a setting where memorization is impossible. Overall, our results show IGDA to be a powerful method for graph discovery complementary to existing numerically driven approaches.
Understanding the effects of economic shocks on firms is critical for analyzing economic growth and resilience. We introduce a Web-Based Affectedness Indicator (WAI), a general-purpose tool for real-time monitoring of economic disruptions across diverse contexts. By leveraging Large Language Model (LLM) assisted classification and information extraction on texts from over five million company websites, WAI quantifies the degree and nature of firms' responses to external shocks. Using the COVID-19 pandemic as a specific application, we show that WAI is highly correlated with pandemic containment measures and reliably predicts firm performance. Unlike traditional data sources, WAI provides timely firm-level information across industries and geographies worldwide that would otherwise be unavailable due to institutional and data availability constraints. This methodology offers significant potential for monitoring and mitigating the impact of technological, political, financial, health or environmental crises, and represents a transformative tool for adaptive policy-making and economic resilience.
Language model (LM) re-rankers are used to refine retrieval results for retrieval-augmented generation (RAG). They are more expensive than lexical matching methods like BM25 but assumed to better process semantic information. To understand whether LM re-rankers always live up to this assumption, we evaluate 6 different LM re-rankers on the NQ, LitQA2 and DRUID datasets. Our results show that LM re-rankers struggle to outperform a simple BM25 re-ranker on DRUID. Leveraging a novel separation metric based on BM25 scores, we explain and identify re-ranker errors stemming from lexical dissimilarities. We also investigate different methods to improve LM re-ranker performance and find these methods mainly useful for NQ. Taken together, our work identifies and explains weaknesses of LM re-rankers and points to the need for more adversarial and realistic datasets for their evaluation.
Understanding the uncertainty in large language model (LLM) explanations is important for evaluating their faithfulness and reasoning consistency, and thus provides insights into the reliability of LLM's output regarding a question. In this work, we propose a novel framework that quantifies uncertainty in LLM explanations through a reasoning topology perspective. By designing a structural elicitation strategy, we guide the LLMs to frame the explanations of an answer into a graph topology. This process decomposes the explanations into the knowledge related sub-questions and topology-based reasoning structures, which allows us to quantify uncertainty not only at the semantic level but also from the reasoning path. It further brings convenience to assess knowledge redundancy and provide interpretable insights into the reasoning process. Our method offers a systematic way to interpret the LLM reasoning, analyze limitations, and provide guidance for enhancing robustness and faithfulness. This work pioneers the use of graph-structured uncertainty measurement in LLM explanations and demonstrates the potential of topology-based quantification.
The growing penetration of renewable energy sources in power systems has increased the complexity and uncertainty of load forecasting, especially for integrated energy systems with multiple energy carriers. Traditional forecasting methods heavily rely on historical data and exhibit limited transferability across different scenarios, posing significant challenges for emerging applications in smart grids and energy internet. This paper proposes the TSLLM-Load Forecasting Mechanism, a novel zero-shot load forecasting framework based on large language models (LLMs) to address these challenges. The framework consists of three key components: a data preprocessing module that handles multi-source energy load data, a time series prompt generation module that bridges the semantic gap between energy data and LLMs through multi-task learning and similarity alignment, and a prediction module that leverages pre-trained LLMs for accurate forecasting. The framework's effectiveness was validated on a real-world dataset comprising load profiles from 20 Australian solar-powered households, demonstrating superior performance in both conventional and zero-shot scenarios. In conventional testing, our method achieved a Mean Squared Error (MSE) of 0.4163 and a Mean Absolute Error (MAE) of 0.3760, outperforming existing approaches by at least 8\%. In zero-shot prediction experiments across 19 households, the framework maintained consistent accuracy with a total MSE of 11.2712 and MAE of 7.6709, showing at least 12\% improvement over current methods. The results validate the framework's potential for accurate and transferable load forecasting in integrated energy systems, particularly beneficial for renewable energy integration and smart grid applications.
Large language models (LLMs) provide a promising way for accurate session-based recommendation (SBR), but they demand substantial computational time and memory. Knowledge distillation (KD)-based methods can alleviate these issues by transferring the knowledge to a small student, which trains a student based on the predictions of a cumbersome teacher. However, these methods encounter difficulties for \textit{LLM-based KD in SBR}. 1) It is expensive to make LLMs predict for all instances in KD. 2) LLMs may make ineffective predictions for some instances in KD, e.g., incorrect predictions for hard instances or similar predictions as existing recommenders for easy instances. In this paper, we propose an active LLM-based KD method in SBR, contributing to sustainable AI. To efficiently distill knowledge from LLMs with limited cost, we propose to extract a small proportion of instances predicted by LLMs. Meanwhile, for a more effective distillation, we propose an active learning strategy to extract instances that are as effective as possible for KD from a theoretical view. Specifically, we first formulate gains based on potential effects (e.g., effective, similar, and incorrect predictions by LLMs) and difficulties (e.g., easy or hard to fit) of instances for KD. Then, we propose to maximize the minimal gains of distillation to find the optimal selection policy for active learning, which can largely avoid extracting ineffective instances in KD. Experiments on real-world datasets show that our method significantly outperforms state-of-the-art methods for SBR.
This paper introduces an enhanced RWKV architecture with adaptive temporal gating mechanisms for improved long-context language modeling. We propose two principal innovations: (1) a position-aware convolutional shift operator that captures local syntactic patterns while preserving global coherence, and (2) a neurally-gated information routing mechanism that dynamically regulates inter-token information flow. Through comprehensive experiments on text generation tasks, our enhanced model demonstrates superior performance compared to the baseline RWKV, achieving 96.5 relative improvement in ROUGE-L scores with only 2.95 increased inference latency. Ablation studies validate the individual contributions of each component, while linguistic analysis reveals the model's adaptive attention to syntactic boundaries and entity coherence. The proposed modifications maintain RWKV's linear computational complexity while significantly enhancing its contextual modeling capabilities, establishing new state-of-the-art performance for recurrent-style architectures in long-form text generation.
This paper presents an effective approach to detect AI-generated text, developed for the Defactify 4.0 shared task at the fourth workshop on multimodal fact checking and hate speech detection. The task consists of two subtasks: Task-A, classifying whether a text is AI generated or human written, and Task-B, classifying the specific large language model that generated the text. Our team (Sarang) achieved the 1st place in both tasks with F1 scores of 1.0 and 0.9531, respectively. The methodology involves adding noise to the dataset to improve model robustness and generalization. We used an ensemble of DeBERTa models to effectively capture complex patterns in the text. The result indicates the effectiveness of our noise-driven and ensemble-based approach, setting a new standard in AI-generated text detection and providing guidance for future developments.
The next token prediction loss is the dominant self-supervised training objective for large language models and has achieved promising results in a variety of downstream tasks. However, upon closer investigation of this objective, we find that it lacks an understanding of sequence-level signals, leading to a mismatch between training and inference processes. To bridge this gap, we introduce a contrastive preference optimization (CPO) procedure that can inject sequence-level information into the language model at any training stage without expensive human labeled data. Our experiments show that the proposed objective surpasses the next token prediction in terms of win rate in the instruction-following and text generation tasks.
Companies across all economic sectors continue to deploy large language models at a rapid pace. Reinforcement learning is experiencing a resurgence of interest due to its association with the fine-tuning of language models from human feedback. Tool-chain language models control task-specific agents; if the converse has not already appeared, it soon will. In this paper, we present what we believe is the first investigation of an intelligent trading agent based on continuous deep reinforcement learning that also controls a large language model with which it can post to a social media feed observed by other traders. We empirically investigate the performance and impact of such an agent in a simulated financial market, finding that it learns to optimize its total reward, and thereby augment its profit, by manipulating the sentiment of the posts it produces. The paper concludes with discussion, limitations, and suggestions for future work.
Generative AI (GenAI) is increasingly used in survey contexts to simulate human preferences. While many research endeavors evaluate the quality of synthetic GenAI data by comparing model-generated responses to gold-standard survey results, fundamental questions about the validity and reliability of using LLMs as substitutes for human respondents remain. Our study provides a technical analysis of how demographic attributes and prompt variations influence latent opinion mappings in large language models (LLMs) and evaluates their suitability for survey-based predictions. Using 14 different models, we find that LLM-generated data fails to replicate the variance observed in real-world human responses, particularly across demographic subgroups. In the political space, persona-to-party mappings exhibit limited differentiation, resulting in synthetic data that lacks the nuanced distribution of opinions found in survey data. Moreover, we show that prompt sensitivity can significantly alter outputs for some models, further undermining the stability and predictiveness of LLM-based simulations. As a key contribution, we adapt a probe-based methodology that reveals how LLMs encode political affiliations in their latent space, exposing the systematic distortions introduced by these models. Our findings highlight critical limitations in AI-generated survey data, urging caution in its use for public opinion research, social science experimentation, and computational behavioral modeling.
Teaching literature under interdisciplinary contexts (e.g., science, art) that connect reading materials has become popular in elementary schools. However, constructing such contexts is challenging as it requires teachers to explore substantial amounts of interdisciplinary content and link it to the reading materials. In this paper, we develop LitLinker via an iterative design process involving 13 teachers to facilitate the ideation of interdisciplinary contexts for teaching literature. Powered by a large language model (LLM), LitLinker can recommend interdisciplinary topics and contextualize them with the literary elements (e.g., paragraphs, viewpoints) in the reading materials. A within-subjects study (N=16) shows that compared to an LLM chatbot, LitLinker can improve the integration depth of different subjects and reduce workload in this ideation task. Expert interviews (N=9) also demonstrate LitLinker's usefulness for supporting the ideation of interdisciplinary contexts for teaching literature. We conclude with concerns and design considerations for supporting interdisciplinary teaching with LLMs.