BENCHAGENTS: Automated Benchmark Creation with Agent Interaction

Evaluations are limited by benchmark availability. As models evolve, there is a need to create benchmarks that can measure progress on new generative capabilities. However, creating new benchmarks through human annotations is slow and expensive, restricting comprehensive evaluations for any capability. We introduce BENCHAGENTS, a framework that methodically leverages large language models (LLMs) to automate benchmark creation for complex capabilities while inherently ensuring data and metric quality. BENCHAGENTS decomposes the benchmark creation process into planning, generation, data verification, and evaluation, each of which is executed by an LLM agent. These agents interact with each other and utilize human-in-the-loop feedback from benchmark developers to explicitly improve and flexibly control data diversity and quality. We use BENCHAGENTS to create benchmarks to evaluate capabilities related to planning and constraint satisfaction during text generation. We then use these benchmarks to study seven state-of-the-art models and extract new insights on common failure modes and model differences.

Attention Speaks Volumes: Localizing and Mitigating Bias in Language Models

We explore the internal mechanisms of how bias emerges in large language models (LLMs) when provided with ambiguous comparative prompts: inputs that compare or enforce choosing between two or more entities without providing clear context for preference. Most approaches for bias mitigation focus on either post-hoc analysis or data augmentation. However, these are transient solutions, without addressing the root cause: the model itself. Numerous prior works show the influence of the attention module towards steering generations. We believe that analyzing attention is also crucial for understanding bias, as it provides insight into how the LLM distributes its focus across different entities and how this contributes to biased decisions. To this end, we first introduce a metric to quantify the LLM's preference for one entity over another. We then propose $\texttt{ATLAS}$ (Attention-based Targeted Layer Analysis and Scaling), a technique to localize bias to specific layers of the LLM by analyzing attention scores and then reduce bias by scaling attention in these biased layers. To evaluate our method, we conduct experiments across 3 datasets (BBQ, Crows-Pairs, and WinoGender) using $\texttt{GPT-2 XL}$ (1.5B), $\texttt{GPT-J}$ (6B), $\texttt{LLaMA-2}$ (7B) and $\texttt{LLaMA-3}$ (8B). Our experiments demonstrate that bias is concentrated in the later layers, typically around the last third. We also show how $\texttt{ATLAS}$ effectively mitigates bias through targeted interventions without compromising downstream performance and an average increase of only 0.82% in perplexity when the intervention is applied. We see an average improvement of 0.28 points in the bias score across all the datasets.

Unearthing Skill-Level Insights for Understanding Trade-Offs of Foundation Models

With models getting stronger, evaluations have grown more complex, testing multiple skills in one benchmark and even in the same instance at once. However, skill-wise performance is obscured when inspecting aggregate accuracy, under-utilizing the rich signal modern benchmarks contain. We propose an automatic approach to recover the underlying skills relevant for any evaluation instance, by way of inspecting model-generated rationales. After validating the relevance of rationale-parsed skills and inferring skills for $46$k instances over $12$ benchmarks, we observe many skills to be common across benchmarks, resulting in the curation of hundreds of skill-slices (i.e. sets of instances testing a common skill). Inspecting accuracy over these slices yields novel insights on model trade-offs: e.g., compared to GPT-4o and Claude 3.5 Sonnet, on average, Gemini 1.5 Pro is $18\%$ more accurate in "computing molar mass", but $19\%$ less accurate in "applying constitutional law", despite the overall accuracies of the three models differing by a mere $0.4\%$. Furthermore, we demonstrate the practical utility of our approach by showing that insights derived from skill slice analysis can generalize to held-out instances: when routing each instance to the model strongest on the relevant skills, we see a $3\%$ accuracy improvement over our $12$ dataset corpus. Our skill-slices and framework open a new avenue in model evaluation, leveraging skill-specific analyses to unlock a more granular and actionable understanding of model capabilities.

Eureka: Evaluating and Understanding Large Foundation Models

Rigorous and reproducible evaluation is critical for assessing the state of the art and for guiding scientific advances in Artificial Intelligence. Evaluation is challenging in practice due to several reasons, including benchmark saturation, lack of transparency in methods used for measurement, development challenges in extracting measurements for generative tasks, and, more generally, the extensive number of capabilities required for a well-rounded comparison across models. We make three contributions to alleviate the above challenges. First, we present Eureka, an open-source framework for standardizing evaluations of large foundation models beyond single-score reporting and rankings. Second, we introduce Eureka-Bench as an extensible collection of benchmarks testing capabilities that (i) are still challenging for state-of-the-art models and (ii) represent fundamental but overlooked language and multimodal capabilities. The inherent space for improvement in non-saturated benchmarks enables us to discover meaningful differences between models at a capability level. Third, using Eureka, we conduct an analysis of 12 state-of-the-art models, providing in-depth insights into failure understanding and model comparison, which can be leveraged to plan targeted improvements. In contrast to recent trends in reports and leaderboards showing absolute rankings and claims for one model or another to be the best, our analysis shows that there is no such best model. Different models have different strengths, but there are models that appear more often than others as best performers for some capabilities. Despite the recent improvements, current models still struggle with several fundamental capabilities including detailed image understanding, benefiting from multimodal input when available rather than fully relying on language, factuality and grounding for information retrieval, and over refusals.

Understanding Information Storage and Transfer in Multi-modal Large Language Models

Understanding the mechanisms of information storage and transfer in Transformer-based models is important for driving model understanding progress. Recent work has studied these mechanisms for Large Language Models (LLMs), revealing insights on how information is stored in a model's parameters and how information flows to and from these parameters in response to specific prompts. However, these studies have not yet been extended to Multi-modal Large Language Models (MLLMs). Given their expanding capabilities and real-world use, we start by studying one aspect of these models -- how MLLMs process information in a factual visual question answering task. We use a constraint-based formulation which views a visual question as having a set of visual or textual constraints that the model's generated answer must satisfy to be correct (e.g. What movie directed by the director in this photo has won a Golden Globe?). Under this setting, we contribute i) a method that extends causal information tracing from pure language to the multi-modal setting, and ii) VQA-Constraints, a test-bed of 9.7K visual questions annotated with constraints. We use these tools to study two open-source MLLMs, LLaVa and multi-modal Phi-2. Our key findings show that these MLLMs rely on MLP and self-attention blocks in much earlier layers for information storage, compared to LLMs whose mid-layer MLPs are more important. We also show that a consistent small subset of visual tokens output by the vision encoder are responsible for transferring information from the image to these causal blocks. We validate these mechanisms by introducing MultEdit, a model-editing algorithm that can correct errors and insert new long-tailed information into MLLMs by targeting these causal blocks.

Introducing v0.5 of the AI Safety Benchmark from MLCommons

This paper introduces v0.5 of the AI Safety Benchmark, which has been created by the MLCommons AI Safety Working Group. The AI Safety Benchmark has been designed to assess the safety risks of AI systems that use chat-tuned language models. We introduce a principled approach to specifying and constructing the benchmark, which for v0.5 covers only a single use case (an adult chatting to a general-purpose assistant in English), and a limited set of personas (i.e., typical users, malicious users, and vulnerable users). We created a new taxonomy of 13 hazard categories, of which 7 have tests in the v0.5 benchmark. We plan to release version 1.0 of the AI Safety Benchmark by the end of 2024. The v1.0 benchmark will provide meaningful insights into the safety of AI systems. However, the v0.5 benchmark should not be used to assess the safety of AI systems. We have sought to fully document the limitations, flaws, and challenges of v0.5. This release of v0.5 of the AI Safety Benchmark includes (1) a principled approach to specifying and constructing the benchmark, which comprises use cases, types of systems under test (SUTs), language and context, personas, tests, and test items; (2) a taxonomy of 13 hazard categories with definitions and subcategories; (3) tests for seven of the hazard categories, each comprising a unique set of test items, i.e., prompts. There are 43,090 test items in total, which we created with templates; (4) a grading system for AI systems against the benchmark; (5) an openly available platform, and downloadable tool, called ModelBench that can be used to evaluate the safety of AI systems on the benchmark; (6) an example evaluation report which benchmarks the performance of over a dozen openly available chat-tuned language models; (7) a test specification for the benchmark.

Elephants Never Forget: Memorization and Learning of Tabular Data in Large Language Models

While many have shown how Large Language Models (LLMs) can be applied to a diverse set of tasks, the critical issues of data contamination and memorization are often glossed over. In this work, we address this concern for tabular data. Specifically, we introduce a variety of different techniques to assess whether a language model has seen a tabular dataset during training. This investigation reveals that LLMs have memorized many popular tabular datasets verbatim. We then compare the few-shot learning performance of LLMs on datasets that were seen during training to the performance on datasets released after training. We find that LLMs perform better on datasets seen during training, indicating that memorization leads to overfitting. At the same time, LLMs show non-trivial performance on novel datasets and are surprisingly robust to data transformations. We then investigate the in-context statistical learning abilities of LLMs. Without fine-tuning, we find them to be limited. This suggests that much of the few-shot performance on novel datasets is due to the LLM's world knowledge. Overall, our results highlight the importance of testing whether an LLM has seen an evaluation dataset during pre-training. We make the exposure tests we developed available as the tabmemcheck Python package at https://github.com/interpretml/LLM-Tabular-Memorization-Checker

KITAB: Evaluating LLMs on Constraint Satisfaction for Information Retrieval

We study the ability of state-of-the art models to answer constraint satisfaction queries for information retrieval (e.g., 'a list of ice cream shops in San Diego'). In the past, such queries were considered to be tasks that could only be solved via web-search or knowledge bases. More recently, large language models (LLMs) have demonstrated initial emergent abilities in this task. However, many current retrieval benchmarks are either saturated or do not measure constraint satisfaction. Motivated by rising concerns around factual incorrectness and hallucinations of LLMs, we present KITAB, a new dataset for measuring constraint satisfaction abilities of language models. KITAB consists of book-related data across more than 600 authors and 13,000 queries, and also offers an associated dynamic data collection and constraint verification approach for acquiring similar test data for other authors. Our extended experiments on GPT4 and GPT3.5 characterize and decouple common failure modes across dimensions such as information popularity, constraint types, and context availability. Results show that in the absence of context, models exhibit severe limitations as measured by irrelevant information, factual errors, and incompleteness, many of which exacerbate as information popularity decreases. While context availability mitigates irrelevant information, it is not helpful for satisfying constraints, identifying fundamental barriers to constraint satisfaction. We open source our contributions to foster further research on improving constraint satisfaction abilities of future models.

Diversity of Thought Improves Reasoning Abilities of Large Language Models

Large language models (LLMs) are documented to struggle in settings that require complex reasoning. Nevertheless, instructing the model to break down the problem into smaller reasoning steps (Wei et al., 2022), or ensembling various generations through modifying decoding steps (Wang et al., 2023) boosts performance. Current methods assume that the input prompt is fixed and expect the decoding strategies to introduce the diversity needed for ensembling. In this work, we relax this assumption and discuss how one can create and leverage variations of the input prompt as a means to diversity of thought to improve model performance. We propose a method that automatically improves prompt diversity by soliciting feedback from the LLM to ideate approaches that fit for the problem. We then ensemble the diverse prompts in our method DIV-SE (DIVerse reasoning path Self-Ensemble) across multiple inference calls. We also propose a cost-effective alternative where diverse prompts are used within a single inference call; we call this IDIV-SE (In-call DIVerse reasoning path Self-Ensemble). Under a fixed generation budget, DIV-SE and IDIV-SE outperform the previously discussed baselines using both GPT-3.5 and GPT-4 on several reasoning benchmarks, without modifying the decoding process. Additionally, DIV-SE advances state-of-the-art performance on recent planning benchmarks (Valmeekam et al., 2023), exceeding the highest previously reported accuracy by at least 29.6 percentage points on the most challenging 4/5 Blocksworld task. Our results shed light on how to enforce prompt diversity toward LLM reasoning and thereby improve the pareto frontier of the accuracy-cost trade-off.

Attention Satisfies: A Constraint-Satisfaction Lens on Factual Errors of Language Models

We investigate the internal behavior of Transformer-based Large Language Models (LLMs) when they generate factually incorrect text. We propose modeling factual queries as Constraint Satisfaction Problems and use this framework to investigate how the model interacts internally with factual constraints. Specifically, we discover a strong positive relation between the model's attention to constraint tokens and the factual accuracy of its responses. In our curated suite of 11 datasets with over 40,000 prompts, we study the task of predicting factual errors with the Llama-2 family across all scales (7B, 13B, 70B). We propose SAT Probe, a method probing self-attention patterns, that can predict constraint satisfaction and factual errors, and allows early error identification. The approach and findings demonstrate how using the mechanistic understanding of factuality in LLMs can enhance reliability.