The rapid development and deployment of Generative AI in social settings raise important questions about how to optimally personalize them for users while maintaining accuracy and realism. Based on a Facebook public post-comment dataset, this study evaluates the ability of Llama 3.0 (70B) to predict the semantic tones across different combinations of a commenter's and poster's gender, age, and friendship closeness and to replicate these differences in LLM-generated comments. The study consists of two parts: Part I assesses differences in semantic tones across social relationship categories, and Part II examines the similarity between comments generated by Llama 3.0 (70B) and human comments from Part I given public Facebook posts as input. Part I results show that including social relationship information improves the ability of a model to predict the semantic tone of human comments. However, Part II results show that even without including social context information in the prompt, LLM-generated comments and human comments are equally sensitive to social context, suggesting that LLMs can comprehend semantics from the original post alone. When we include all social relationship information in the prompt, the similarity between human comments and LLM-generated comments decreases. This inconsistency may occur because LLMs did not include social context information as part of their training data. Together these results demonstrate the ability of LLMs to comprehend semantics from the original post and respond similarly to human comments, but also highlights their limitations in generalizing personalized comments through prompting alone.

State-of-the-art language models can exhibit impressive reasoning refinement capabilities on math, science or coding tasks. However, recent work demonstrates that even the best models struggle to identify \textit{when and where to refine} without access to external feedback. Outcome-based Reward Models (\textbf{ORMs}), trained to predict correctness of the final answer indicating when to refine, offer one convenient solution for deciding when to refine. Process Based Reward Models (\textbf{PRMs}), trained to predict correctness of intermediate steps, can then be used to indicate where to refine. But they are expensive to train, requiring extensive human annotations. In this paper, we propose Stepwise ORMs (\textbf{SORMs}) which are trained, only on synthetic data, to approximate the expected future reward of the optimal policy or $V^{\star}$. More specifically, SORMs are trained to predict the correctness of the final answer when sampling the current policy many times (rather than only once as in the case of ORMs). Our experiments show that SORMs can more accurately detect incorrect reasoning steps compared to ORMs, thus improving downstream accuracy when doing refinements. We then train \textit{global} refinement models, which take only the question and a draft solution as input and predict a corrected solution, and \textit{local} refinement models which also take as input a critique indicating the location of the first reasoning error. We generate training data for both models synthetically by reusing data used to train the SORM. We find combining global and local refinements, using the ORM as a reranker, significantly outperforms either one individually, as well as a best of three sample baseline. With this strategy we can improve the accuracy of a LLaMA-2 13B model (already fine-tuned with RL) on GSM8K from 53\% to 65\% when greedily sampled.

Information Retrieval (IR) systems are designed to deliver relevant content, but traditional systems may not optimize rankings for fairness, neutrality, or the balance of ideas. Consequently, IR can often introduce indexical biases, or biases in the positional order of documents. Although indexical bias can demonstrably affect people's opinion, voting patterns, and other behaviors, these issues remain understudied as the field lacks reliable metrics and procedures for automatically measuring indexical bias. Towards this end, we introduce the PAIR framework, which supports automatic bias audits for ranked documents or entire IR systems. After introducing DUO, the first general-purpose automatic bias metric, we run an extensive evaluation of 8 IR systems on a new corpus of 32k synthetic and 4.7k natural documents, with 4k queries spanning 1.4k controversial issue topics. A human behavioral study validates our approach, showing that our bias metric can help predict when and how indexical bias will shift a reader's opinion.

The accurate evaluation of differential treatment in language models to specific groups is critical to ensuring a positive and safe user experience. An ideal evaluation should have the properties of being robust, extendable to new groups or attributes, and being able to capture biases that appear in typical usage (rather than just extreme, rare cases). Relatedly, bias evaluation should surface not only egregious biases but also ones that are subtle and commonplace, such as a likelihood for talking about appearances with regard to women. We present FairPair, an evaluation framework for assessing differential treatment that occurs during ordinary usage. FairPair operates through counterfactual pairs, but crucially, the paired continuations are grounded in the same demographic group, which ensures equivalent comparison. Additionally, unlike prior work, our method factors in the inherent variability that comes from the generation process itself by measuring the sampling variability. We present an evaluation of several commonly used generative models and a qualitative analysis that indicates a preference for discussing family and hobbies with regard to women.

Teaching language models to use tools is an important milestone towards building general assistants, but remains an open problem. While there has been significant progress on learning to use specific tools via fine-tuning, language models still struggle with learning how to robustly use new tools from only a few demonstrations. In this work we introduce a self-verification method which distinguishes between close candidates by self-asking contrastive questions during (1) tool selection; and (2) parameter generation. We construct synthetic, high-quality, self-generated data for this goal using Llama-2 70B, which we intend to release publicly. Extensive experiments on 4 tasks from the ToolBench benchmark, consisting of 17 unseen tools, demonstrate an average improvement of 22% over few-shot baselines, even in scenarios where the distinctions between candidate tools are finely nuanced.

Simultaneously, Reinforcement Learning from Human Feedback (RLHF) (Bai et al., 2022; Ziegler et al., 2019; Ouyang et al., 2022) and instruction fine-tuning (Wei et al., 2021; Mishra et al., 2021) have made significant progress in aligning LLMs with human preferences. Improvements in model instructability have further increased apparent model capability by making complex behaviors more accessible via instruction prompting. This has led to a number of increasingly sophisticated prompting strategies augmenting LLM reasoning capabilities such as Chain-of-Thought (Wei et al., 2022) or Tree-of-Thoughts (Yao et al., 2023). Previous work in reinforcement learning (RL) such as AlphaGo (Silver et al., 2017), AlphaStar (Vinyals et al., 2019), and OpenAI Dota 2 (Berner et al., 2019) demonstrate that RL techniques can be used to train neural networks capable of sophisticated planning and reasoning in game environments. Cicero (Bakhtin et al., 2022) in particular succeeds in combining an RL trained planning agent with a dialogue fine-tuned LLM to achieve nearly super-human performance in the board game Diplomacy. Given these previous successes and the inherent interactive nature of problem solving, applying RL to LLM reasoning seems a natural next step. In this paper, we study how ideas from RL can be used to improve the reasoning capabilities of LLMs across a variety of reward schemes and model initializations. We begin by comparing the performance of different RL algorithms on reasoning tasks τ defined as a distribution of question answer tuples (Q, A).

Dense retrievers compress source documents into (possibly lossy) vector representations, yet there is little analysis of what information is lost versus preserved, and how it affects downstream tasks. We conduct the first analysis of the information captured by dense retrievers compared to the language models they are based on (e.g., BERT versus Contriever). We use 25 MultiBert checkpoints as randomized initialisations to train MultiContrievers, a set of 25 contriever models. We test whether specific pieces of information -- such as gender and occupation -- can be extracted from contriever vectors of wikipedia-like documents. We measure this extractability via information theoretic probing. We then examine the relationship of extractability to performance and gender bias, as well as the sensitivity of these results to many random initialisations and data shuffles. We find that (1) contriever models have significantly increased extractability, but extractability usually correlates poorly with benchmark performance 2) gender bias is present, but is not caused by the contriever representations 3) there is high sensitivity to both random initialisation and to data shuffle, suggesting that future retrieval research should test across a wider spread of both.

To achieve faithful reasoning that aligns with human expectations, large language models (LLMs) need to ground their reasoning to real-world knowledge (e.g., web facts, math and physical rules). Tools help LLMs access this external knowledge, but there remains challenges for fine-tuning LLM agents (e.g., Toolformer) to invoke tools in multi-step reasoning problems, where inter-connected tool calls require holistic and efficient tool usage planning. In this work, we propose a new method for LLMs to better leverage tools in multi-step reasoning. Our method, Chain-of-Abstraction (CoA), trains LLMs to first decode reasoning chains with abstract placeholders, and then call domain tools to reify each reasoning chain by filling in specific knowledge. This planning with abstract chains enables LLMs to learn more general reasoning strategies, which are robust to shifts of domain knowledge (e.g., math results) relevant to different reasoning questions. It also allows LLMs to perform decoding and calling of external tools in parallel, which avoids the inference delay caused by waiting for tool responses. In mathematical reasoning and Wiki QA domains, we show that our method consistently outperforms previous chain-of-thought and tool-augmented baselines on both in-distribution and out-of-distribution test sets, with an average ~6% absolute QA accuracy improvement. LLM agents trained with our method also show more efficient tool use, with inference speed being on average ~1.4x faster than baseline tool-augmented LLMs.

As generative large language models (LLMs) grow more performant and prevalent, we must develop comprehensive enough tools to measure and improve their fairness. Different prompt-based datasets can be used to measure social bias across multiple text domains and demographic axes, meaning that testing LLMs on more datasets can potentially help us characterize their biases more fully, and better ensure equal and equitable treatment of marginalized demographic groups. In this work, our focus is two-fold: (1) Benchmarking: a comparison of 6 different prompt-based bias and toxicity metrics across 12 demographic axes and 5 families of generative LLMs. Out of those 6 metrics, AdvPromptSet and HolisticBiasR are novel datasets proposed in the paper. The comparison of those benchmarks gives us insights about the bias and toxicity of the compared models. Therefore, we explore the frequency of demographic terms in common LLM pre-training corpora and how this may relate to model biases. (2) Mitigation: we conduct a comprehensive study of how well 3 bias/toxicity mitigation techniques perform across our suite of measurements. ROBBIE aims to provide insights for practitioners while deploying a model, emphasizing the need to not only measure potential harms, but also understand how they arise by characterizing the data, mitigate harms once found, and balance any trade-offs. We open-source our analysis code in hopes of encouraging broader measurements of bias in future LLMs.

The remarkable advancements in large language models (LLMs) have significantly enhanced the performance in few-shot learning settings. By using only a small number of labeled examples, referred to as demonstrations, LLMs can effectively grasp the task at hand through in-context learning. However, the process of selecting appropriate demonstrations has received limited attention in prior work. This paper addresses the issue of identifying the most informative demonstrations for few-shot learning by approaching it as a pool-based Active Learning (AL) problem over a single iteration. Our objective is to investigate how AL algorithms can serve as effective demonstration selection methods for in-context learning. We compare various standard AL algorithms based on uncertainty, diversity, and similarity, and consistently observe that the latter outperforms all other methods, including random sampling. Notably, uncertainty sampling, despite its success in conventional supervised learning scenarios, performs poorly in this context. Our extensive experimentation involving a diverse range of GPT and OPT models across $24$ classification and multi-choice tasks, coupled with thorough analysis, unambiguously demonstrates that in-context example selection through AL prioritizes high-quality examples that exhibit low uncertainty and bear similarity to the test examples.