Reinforcement learning algorithms typically struggle in the absence of a dense, well-shaped reward function. Intrinsically motivated exploration methods address this limitation by rewarding agents for visiting novel states or transitions, but these methods offer limited benefits in large environments where most discovered novelty is irrelevant for downstream tasks. We describe a method that uses background knowledge from text corpora to shape exploration. This method, called ELLM (Exploring with LLMs) rewards an agent for achieving goals suggested by a language model prompted with a description of the agent's current state. By leveraging large-scale language model pretraining, ELLM guides agents toward human-meaningful and plausibly useful behaviors without requiring a human in the loop. We evaluate ELLM in the Crafter game environment and the Housekeep robotic simulator, showing that ELLM-trained agents have better coverage of common-sense behaviors during pretraining and usually match or improve performance on a range of downstream tasks. Code available at https://github.com/yuqingd/ellm.

In the field of artificial intelligence (AI), the quest to understand and model data-generating processes (DGPs) is of paramount importance. Deep generative models (DGMs) have proven adept in capturing complex data distributions but often fall short in generalization and interpretability. On the other hand, causality offers a structured lens to comprehend the mechanisms driving data generation and highlights the causal-effect dynamics inherent in these processes. While causality excels in interpretability and the ability to extrapolate, it grapples with intricacies of high-dimensional spaces. Recognizing the synergistic potential, we delve into the confluence of causality and DGMs. We elucidate the integration of causal principles within DGMs, investigate causal identification using DGMs, and navigate an emerging research frontier of causality in large-scale generative models, particularly generative large language models (LLMs). We offer insights into methodologies, highlight open challenges, and suggest future directions, positioning our comprehensive review as an essential guide in this swiftly emerging and evolving area.

We present a new financial domain large language model, InvestLM, tuned on LLaMA-65B (Touvron et al., 2023), using a carefully curated instruction dataset related to financial investment. Inspired by less-is-more-for-alignment (Zhou et al., 2023), we manually curate a small yet diverse instruction dataset, covering a wide range of financial related topics, from Chartered Financial Analyst (CFA) exam questions to SEC filings to Stackexchange quantitative finance discussions. InvestLM shows strong capabilities in understanding financial text and provides helpful responses to investment related questions. Financial experts, including hedge fund managers and research analysts, rate InvestLM's response as comparable to those of state-of-the-art commercial models (GPT-3.5, GPT-4 and Claude-2). Zero-shot evaluation on a set of financial NLP benchmarks demonstrates strong generalizability. From a research perspective, this work suggests that a high-quality domain specific LLM can be tuned using a small set of carefully curated instructions on a well-trained foundation model, which is consistent with the Superficial Alignment Hypothesis (Zhou et al., 2023). From a practical perspective, this work develops a state-of-the-art financial domain LLM with superior capability in understanding financial texts and providing helpful investment advice, potentially enhancing the work efficiency of financial professionals. We release the model parameters to the research community.

Modern language models often exhibit powerful but brittle behavior, leading to the development of larger and more diverse benchmarks to reliably assess their behavior. Here, we suggest that model performance can be benchmarked and elucidated with much smaller evaluation sets. We first show that in six popular language classification benchmarks, model confidence in the correct class on many pairs of points is strongly correlated across models. We build upon this phenomenon to propose Anchor Point Selection, a technique to select small subsets of datasets that capture model behavior across the entire dataset. Anchor points reliably rank models: across 87 diverse language model-prompt pairs, evaluating models using 1-30 anchor points outperforms uniform sampling and other baselines at accurately ranking models. Moreover, just several anchor points can be used to estimate model per-class predictions on all other points in a dataset with low mean absolute error, sufficient for gauging where the model is likely to fail. Lastly, we present Anchor Point Maps for visualizing these insights and facilitating comparisons of the performance of different models on various regions within the dataset distribution.

Summarization is an important application of large language models (LLMs). Most previous evaluation of summarization models has focused on their performance in content selection, grammaticality and coherence. However, it is well known that LLMs reproduce and reinforce harmful social biases. This raises the question: Do these biases affect model outputs in a relatively constrained setting like summarization? To help answer this question, we first motivate and introduce a number of definitions for biased behaviours in summarization models, along with practical measures to quantify them. Since we find biases inherent to the input document can confound our analysis, we additionally propose a method to generate input documents with carefully controlled demographic attributes. This allows us to sidestep this issue, while still working with somewhat realistic input documents. Finally, we apply our measures to summaries generated by both purpose-built summarization models and general purpose chat models. We find that content selection in single document summarization seems to be largely unaffected by bias, while hallucinations exhibit evidence of biases propagating to generated summaries.

Large language models (LLMs) have shown remarkable capabilities in Natural Language Processing (NLP), especially in domains where labeled data is scarce or expensive, such as clinical domain. However, to unlock the clinical knowledge hidden in these LLMs, we need to design effective prompts that can guide them to perform specific clinical NLP tasks without any task-specific training data. This is known as in-context learning, which is an art and science that requires understanding the strengths and weaknesses of different LLMs and prompt engineering approaches. In this paper, we present a comprehensive and systematic experimental study on prompt engineering for five clinical NLP tasks: Clinical Sense Disambiguation, Biomedical Evidence Extraction, Coreference Resolution, Medication Status Extraction, and Medication Attribute Extraction. We assessed the prompts proposed in recent literature, including simple prefix, simple cloze, chain of thought, and anticipatory prompts, and introduced two new types of prompts, namely heuristic prompting and ensemble prompting. We evaluated the performance of these prompts on three state-of-the-art LLMs: GPT-3.5, BARD, and LLAMA2. We also contrasted zero-shot prompting with few-shot prompting, and provide novel insights and guidelines for prompt engineering for LLMs in clinical NLP. To the best of our knowledge, this is one of the first works on the empirical evaluation of different prompt engineering approaches for clinical NLP in this era of generative AI, and we hope that it will inspire and inform future research in this area.

Recent progress in using machine learning models for reasoning tasks has been driven by novel model architectures, large-scale pre-training protocols, and dedicated reasoning datasets for fine-tuning. In this work, to further pursue these advances, we introduce a new data generator for machine reasoning that integrates with an embodied agent. The generated data consists of templated text queries and answers, matched with world-states encoded into a database. The world-states are a result of both world dynamics and the actions of the agent. We show the results of several baseline models on instantiations of train sets. These include pre-trained language models fine-tuned on a text-formatted representation of the database, and graph-structured Transformers operating on a knowledge-graph representation of the database. We find that these models can answer some questions about the world-state, but struggle with others. These results hint at new research directions in designing neural reasoning models and database representations. Code to generate the data will be released at github.com/facebookresearch/neuralmemory

As language models are adapted by a more sophisticated and diverse set of users, the importance of guaranteeing that they provide factually correct information supported by verifiable sources is critical across fields of study & professions. This is especially the case for high-stakes fields, such as medicine and law, where the risk of propagating false information is high and can lead to undesirable societal consequences. Previous work studying factuality and attribution has not focused on analyzing these characteristics of language model outputs in domain-specific scenarios. In this work, we present an evaluation study analyzing various axes of factuality and attribution provided in responses from a few systems, by bringing domain experts in the loop. Specifically, we first collect expert-curated questions from 484 participants across 32 fields of study, and then ask the same experts to evaluate generated responses to their own questions. We also ask experts to revise answers produced by language models, which leads to ExpertQA, a high-quality long-form QA dataset with 2177 questions spanning 32 fields, along with verified answers and attributions for claims in the answers.

Due to the modern relevance of blockchain technology, smart contracts present both substantial risks and benefits. Vulnerabilities within them can trigger a cascade of consequences, resulting in significant losses. Many current papers primarily focus on classifying smart contracts for malicious intent, often relying on limited contract characteristics, such as bytecode or opcode. This paper proposes a novel, two-layered framework: 1) classifying and 2) directly repairing malicious contracts. Slither's vulnerability report is combined with source code and passed through a pre-trained RandomForestClassifier (RFC) and Large Language Models (LLMs), classifying and repairing each suggested vulnerability. Experiments demonstrate the effectiveness of fine-tuned and prompt-engineered LLMs. The smart contract repair models, built from pre-trained GPT-3.5-Turbo and fine-tuned Llama-2-7B models, reduced the overall vulnerability count by 97.5% and 96.7% respectively. A manual inspection of repaired contracts shows that all retain functionality, indicating that the proposed method is appropriate for automatic batch classification and repair of vulnerabilities in smart contracts.

Large Language Models (LLMs) have shown impressive performance across a variety of Artificial Intelligence (AI) and natural language processing tasks, such as content creation, report generation, etc. However, unregulated malign application of these models can create undesirable consequences such as generation of fake news, plagiarism, etc. As a result, accurate detection of AI-generated language can be crucial in responsible usage of LLMs. In this work, we explore 1) whether a certain body of text is AI generated or written by human, and 2) attribution of a specific language model in generating a body of text. Texts in both English and Spanish are considered. The datasets used in this study are provided as part of the Automated Text Identification (AuTexTification) shared task. For each of the research objectives stated above, we propose an ensemble neural model that generates probabilities from different pre-trained LLMs which are used as features to a Traditional Machine Learning (TML) classifier following it. For the first task of distinguishing between AI and human generated text, our model ranked in fifth and thirteenth place (with macro $F1$ scores of 0.733 and 0.649) for English and Spanish texts, respectively. For the second task on model attribution, our model ranked in first place with macro $F1$ scores of 0.625 and 0.653 for English and Spanish texts, respectively.