Recent research highlights the significant potential of ChatGPT for text annotation in social science research. However, ChatGPT is a closed-source product which has major drawbacks with regards to transparency, reproducibility, cost, and data protection. Recent advances in open-source (OS) large language models (LLMs) offer alternatives which remedy these challenges. This means that it is important to evaluate the performance of OS LLMs relative to ChatGPT and standard approaches to supervised machine learning classification. We conduct a systematic comparative evaluation of the performance of a range of OS LLM models alongside ChatGPT, using both zero- and few-shot learning as well as generic and custom prompts, with results compared to more traditional supervised classification models. Using a new dataset of Tweets from US news media, and focusing on simple binary text annotation tasks for standard social science concepts, we find significant variation in the performance of ChatGPT and OS models across the tasks, and that supervised classifiers consistently outperform both. Given the unreliable performance of ChatGPT and the significant challenges it poses to Open Science we advise against using ChatGPT for substantive text annotation tasks in social science research.

Deception and persuasion play a critical role in long-horizon dialogues between multiple parties, especially when the interests, goals, and motivations of the participants are not aligned. Such complex tasks pose challenges for current Large Language Models (LLM) as deception and persuasion can easily mislead them, especially in long-horizon multi-party dialogues. To this end, we explore the game of Avalon: The Resistance, a social deduction game in which players must determine each other's hidden identities to complete their team's objective. We introduce an online testbed and a dataset containing 20 carefully collected and labeled games among human players that exhibit long-horizon deception in a cooperative-competitive setting. We discuss the capabilities of LLMs to utilize deceptive long-horizon conversations between six human players to determine each player's goal and motivation. Particularly, we discuss the multimodal integration of the chat between the players and the game's state that grounds the conversation, providing further insights into the true player identities. We find that even current state-of-the-art LLMs do not reach human performance, making our dataset a compelling benchmark to investigate the decision-making and language-processing capabilities of LLMs. Our dataset and online testbed can be found at our project website: https://sstepput.github.io/Avalon-NLU/

Prompt engineering is a challenging yet crucial task for optimizing the performance of large language models (LLMs). It requires complex reasoning to examine the model's errors, hypothesize what is missing or misleading in the current prompt, and communicate the task with clarity. While recent works indicate that LLMs can be meta-prompted to perform automatic prompt engineering, their potentials may not be fully untapped due to the lack of sufficient guidance to elicit complex reasoning capabilities in LLMs in the meta-prompt. In this work, we investigate the problem of "prompt engineering a prompt engineer" -- constructing a meta-prompt that more effectively guides LLMs to perform automatic prompt engineering. We introduce and analyze key components, such as a step-by-step reasoning template and context specification, which lead to improved performance. In addition, inspired by common optimization concepts such as batch size, step size and momentum, we introduce their verbalized counterparts to the meta-prompt and investigate their effects. Our final method, named PE2, finds a prompt that outperforms "let's think step by step" by 6.3% on the MultiArith dataset and 3.1% on the GSM8K dataset. To demonstrate its versatility, we apply PE2 to the Instruction Induction benchmark, a suite of counterfactual tasks, and a lengthy, real-world industrial prompt. In these settings, PE2 achieves strong performance and outperforms prior automatic prompt engineering baselines. Further, we show that PE2 makes meaningful and targeted prompt edits, amends erroneous or incomplete prompts, and presents non-trivial counterfactual reasoning abilities.

Understanding event descriptions is a central aspect of language processing, but current approaches focus overwhelmingly on single sentences or documents. Aggregating information about an event \emph{across documents} can offer a much richer understanding. To this end, we present FAMuS, a new corpus of Wikipedia passages that \emph{report} on some event, paired with underlying, genre-diverse (non-Wikipedia) \emph{source} articles for the same event. Events and (cross-sentence) arguments in both report and source are annotated against FrameNet, providing broad coverage of different event types. We present results on two key event understanding tasks enabled by FAMuS: \emph{source validation} -- determining whether a document is a valid source for a target report event -- and \emph{cross-document argument extraction} -- full-document argument extraction for a target event from both its report and the correct source article. We release both FAMuS and our models to support further research.

Solving long-horizon, temporally-extended tasks using Reinforcement Learning (RL) is challenging, compounded by the common practice of learning without prior knowledge (or tabula rasa learning). Humans can generate and execute plans with temporally-extended actions and quickly learn to perform new tasks because we almost never solve problems from scratch. We want autonomous agents to have this same ability. Recently, LLMs have been shown to encode a tremendous amount of knowledge about the world and to perform impressive in-context learning and reasoning. However, using LLMs to solve real world problems is hard because they are not grounded in the current task. In this paper we exploit the planning capabilities of LLMs while using RL to provide learning from the environment, resulting in a hierarchical agent that uses LLMs to solve long-horizon tasks. Instead of completely relying on LLMs, they guide a high-level policy, making learning significantly more sample efficient. This approach is evaluated in simulation environments such as MiniGrid, SkillHack, and Crafter, and on a real robot arm in block manipulation tasks. We show that agents trained using our approach outperform other baselines methods and, once trained, don't need access to LLMs during deployment.

Background: General-purpose large language models that utilize both text and images have not been evaluated on a diverse array of challenging medical cases. Methods: Using 934 cases from the NEJM Image Challenge published between 2005 and 2023, we evaluated the accuracy of the recently released Generative Pre-trained Transformer 4 with Vision model (GPT-4V) compared to human respondents overall and stratified by question difficulty, image type, and skin tone. We further conducted a physician evaluation of GPT-4V on 69 NEJM clinicopathological conferences (CPCs). Analyses were conducted for models utilizing text alone, images alone, and both text and images. Results: GPT-4V achieved an overall accuracy of 61% (95% CI, 58 to 64%) compared to 49% (95% CI, 49 to 50%) for humans. GPT-4V outperformed humans at all levels of difficulty and disagreement, skin tones, and image types; the exception was radiographic images, where performance was equivalent between GPT-4V and human respondents. Longer, more informative captions were associated with improved performance for GPT-4V but similar performance for human respondents. GPT-4V included the correct diagnosis in its differential for 80% (95% CI, 68 to 88%) of CPCs when using text alone, compared to 58% (95% CI, 45 to 70%) of CPCs when using both images and text. Conclusions: GPT-4V outperformed human respondents on challenging medical cases and was able to synthesize information from both images and text, but performance deteriorated when images were added to highly informative text. Overall, our results suggest that multimodal AI models may be useful in medical diagnostic reasoning but that their accuracy may depend heavily on context.

Large language models (LLMs) have emerged as powerful and general solutions to many natural language tasks. However, many of the most important applications of language generation are interactive, where an agent has to talk to a person to reach a desired outcome. For example, a teacher might try to understand their student's current comprehension level to tailor their instruction accordingly, and a travel agent might ask questions of their customer to understand their preferences in order to recommend activities they might enjoy. LLMs trained with supervised fine-tuning or "single-step" RL, as with standard RLHF, might struggle which tasks that require such goal-directed behavior, since they are not trained to optimize for overall conversational outcomes after multiple turns of interaction. In this work, we explore a new method for adapting LLMs with RL for such goal-directed dialogue. Our key insight is that, though LLMs might not effectively solve goal-directed dialogue tasks out of the box, they can provide useful data for solving such tasks by simulating suboptimal but human-like behaviors. Given a textual description of a goal-directed dialogue task, we leverage LLMs to sample diverse synthetic rollouts of hypothetical in-domain human-human interactions. Our algorithm then utilizes this dataset with offline reinforcement learning to train an interactive conversational agent that can optimize goal-directed objectives over multiple turns. In effect, the LLM produces examples of possible interactions, and RL then processes these examples to learn to perform more optimal interactions. Empirically, we show that our proposed approach achieves state-of-the-art performance in various goal-directed dialogue tasks that include teaching and preference elicitation.

Fairness in Language Models (LMs) remains a longstanding challenge, given the inherent biases in training data that can be perpetuated by models and affect the downstream tasks. Recent methods employ expensive retraining or attempt debiasing during inference by constraining model outputs to contrast from a reference set of biased templates or exemplars. Regardless, they dont address the primary goal of fairness to maintain equitability across different demographic groups. In this work, we posit that inferencing LMs to generate unbiased output for one demographic under a context ensues from being aware of outputs for other demographics under the same context. To this end, we propose Counterfactually Aware Fair InferencE (CAFIE), a framework that dynamically compares the model understanding of diverse demographics to generate more equitable sentences. We conduct an extensive empirical evaluation using base LMs of varying sizes and across three diverse datasets and found that CAFIE outperforms strong baselines. CAFIE produces fairer text and strikes the best balance between fairness and language modeling capability

Current conversational agents (CA) have seen improvement in conversational quality in recent years due to the influence of large language models (LLMs) like GPT3. However, two key categories of problem remain. Firstly there are the unique technical problems resulting from the approach taken in creating the CA, such as scope with retrieval agents and the often nonsensical answers of former generative agents. Secondly, humans perceive CAs as social actors, and as a result expect the CA to adhere to social convention. Failure on the part of the CA in this respect can lead to a poor interaction and even the perception of threat by the user. As such, this paper presents a survey highlighting a potential solution to both categories of problem through the introduction of cognitively inspired additions to the CA. Through computational facsimiles of semantic and episodic memory, emotion, working memory, and the ability to learn, it is possible to address both the technical and social problems encountered by CAs.

With large language models (LLMs) appearing to behave increasingly human-like in text-based interactions, it has become popular to attempt to evaluate various properties of these models using tests originally designed for humans. While re-using existing tests is a resource-efficient way to evaluate LLMs, careful adjustments are usually required to ensure that test results are even valid across human sub-populations. Thus, it is not clear to what extent different tests' validity generalizes to LLMs. In this work, we provide evidence that LLMs' responses to personality tests systematically deviate from typical human responses, implying that these results cannot be interpreted in the same way as human test results. Concretely, reverse-coded items (e.g. "I am introverted" vs "I am extraverted") are often both answered affirmatively by LLMs. In addition, variation across different prompts designed to "steer" LLMs to simulate particular personality types does not follow the clear separation into five independent personality factors from human samples. In light of these results, we believe it is important to pay more attention to tests' validity for LLMs before drawing strong conclusions about potentially ill-defined concepts like LLMs' "personality".