Contemporary Large Language Models (LLMs) exhibit a high degree of code generation and comprehension capability. A particularly promising area is their ability to interpret code modules from unfamiliar libraries for solving user-instructed tasks. Recent work has shown that large proprietary LLMs can learn novel library usage in-context from demonstrations. These results raise several open questions: whether demonstrations of library usage is required, whether smaller (and more open) models also possess such capabilities, etc. In this work, we take a broader approach by systematically evaluating a diverse array of LLMs across three scenarios reflecting varying levels of domain specialization to understand their abilities and limitations in generating code based on libraries defined in-context. Our results show that even smaller open-source LLMs like Llama-2 and StarCoder demonstrate an adept understanding of novel code libraries based on specification presented in-context. Our findings further reveal that LLMs exhibit a surprisingly high proficiency in learning novel library modules even when provided with just natural language descriptions or raw code implementations of the functions, which are often cheaper to obtain than demonstrations. Overall, our results pave the way for harnessing LLMs in more adaptable and dynamic coding environments.

Large Language Models (LLMs) serve as repositories of extensive world knowledge, enabling them to perform tasks such as question-answering and fact-checking. However, this knowledge can become obsolete as global contexts change. In this paper, we introduce a novel problem in the realm of continual learning: Online Continual Knowledge Learning (OCKL). This problem formulation aims to manage the dynamic nature of world knowledge in LMs under real-time constraints. We propose a new benchmark and evaluation metric designed to measure both the rate of new knowledge acquisition and the retention of previously learned knowledge. Our empirical evaluation, conducted using a variety of state-of-the-art methods, establishes robust base-lines for OCKL. Our results reveal that existing continual learning approaches are unfortunately insufficient for tackling the unique challenges posed by OCKL. We identify key factors that influence the trade-off between knowledge acquisition and retention, thereby advancing our understanding of how to train LMs in a continually evolving environment.

Large language models (LLMs) executing tasks through instruction-based prompts often face challenges stemming from distribution differences between user instructions and training instructions. This leads to distractions and biases, especially when dealing with inconsistent dynamic labels. In this paper, we introduces a novel bias mitigation method, CRISPR, designed to alleviate instruction-label biases in LLMs. CRISPR utilizes attribution methods to identify bias neurons influencing biased outputs and employs pruning to eliminate the bias neurons. Experimental results demonstrate the method's effectiveness in mitigating biases in instruction-based prompting, enhancing language model performance on social bias benchmarks without compromising pre-existing knowledge. CRISPR proves highly practical, model-agnostic, offering flexibility in adapting to evolving social biases.

In-Context Learning (ICL) is an emergent capability of Large Language Models (LLMs). Only a few demonstrations enable LLMs to be used as blackbox for new tasks. Previous studies have shown that using LLMs' outputs as labels is effective in training models to select demonstrations. Such a label is expected to estimate utility of a demonstration in ICL; however, it has not been well understood how different labeling strategies affect results on target tasks. This paper presents an analysis on different utility functions by focusing on LLMs' output probability given ground-truth output, and task-specific reward given LLMs' prediction. Unlike the previous work, we introduce a novel labeling method, incremental utility, which estimates how much incremental knowledge is brought into the LLMs by a demonstration. We conduct experiments with instruction-tuned LLMs on binary/multi-class classification, segmentation, and translation across Arabic, English, Finnish, Japanese, and Spanish. Our results show that (1) the probability is effective when the probability values are distributed across the whole value range (on the classification tasks), and (2) the downstream metric is more robust when nuanced reward values are provided with long outputs (on the segmentation and translation tasks). We then show that the proposed incremental utility further helps ICL by contrasting how the LLMs perform with and without the demonstrations.

Accurately simulating human opinion dynamics is crucial for understanding a variety of societal phenomena, including polarization and the spread of misinformation. However, the agent-based models (ABMs) commonly used for such simulations lack fidelity to human behavior. We propose a new approach to simulating opinion dynamics based on populations of Large Language Models (LLMs). Our findings reveal a strong inherent bias in LLM agents towards accurate information, leading to consensus in line with scientific reality. However, this bias limits the simulation of individuals with resistant views on issues like climate change. After inducing confirmation bias through prompt engineering, we observed opinion fragmentation in line with existing agent-based research. These insights highlight the promise and limitations of LLM agents in this domain and suggest a path forward: refining LLMs with real-world discourse to better simulate the evolution of human beliefs.

Augmenting a language model (LM) with $k$-nearest neighbors (kNN) retrieval on its training data alone can decrease its perplexity, though the underlying reasons for this remains elusive. In this work, we first rule out one previously posited possibility -- the "softmax bottleneck." We further identify the MLP hurdle phenomenon, where the final MLP layer in LMs may impede LM optimization early on. We explore memorization and generalization in language models with two new datasets, where advanced model like GPT-3.5-turbo find generalizing to irrelevant information in the training data challenging. However, incorporating kNN retrieval to vanilla GPT-2 117M can consistently improve performance in this setting.

While LLMs can provide reasoned explanations along with their answers, the nature and quality of those explanations are still poorly understood. In response, our goal is to define a detailed way of characterizing the explanation capabilities of modern models and to create a nuanced, interpretable explanation evaluation tool that can generate such characterizations automatically, without relying on expensive API calls or human annotations. Our approach is to (a) define the new task of explanation critiquing - identifying and categorizing any main flaw in an explanation and providing suggestions to address the flaw, (b) create a sizeable, human-verified dataset for this task, and (c) train an open-source, automatic critiquing model (called Digital Socrates) using this data. Through quantitative and qualitative analysis, we demonstrate how Digital Socrates is useful for revealing insights about student models by examining their reasoning chains, and how it can provide high-quality, nuanced, automatic evaluation of those model explanations for the first time. Digital Socrates thus fills an important gap in evaluation tools for understanding and improving the explanation behavior of models.

Large language models (LLMs) have the ability to perform in-context learning (ICL) of new tasks by conditioning on prompts comprising a few task examples. This work studies the problem of selecting the best examples given a candidate pool to improve ICL performance on given a test input. Existing approaches either require training with feedback from a much larger LLM or are computationally expensive. We propose a novel metric, GistScore, based on Example Gisting, a novel approach for training example retrievers for ICL using an attention bottleneck via Gisting, a recent technique for compressing task instructions. To tradeoff performance with ease of use, we experiment with both fine-tuning gist models on each dataset and multi-task training a single model on a large collection of datasets. On 21 diverse datasets spanning 9 tasks, we show that our fine-tuned models get state-of-the-art ICL performance with 20% absolute average gain over off-the-shelf retrievers and 7% over the best prior methods. Our multi-task model generalizes well out-of-the-box to new task categories, datasets, and prompt templates with retrieval speeds that are consistently thousands of times faster than the best prior training-free method.

The inevitable appearance of spurious correlations in training datasets hurts the generalization of NLP models on unseen data. Previous work has found that datasets with paired inputs are prone to correlations between a specific part of the input (e.g., the hypothesis in NLI) and the label; consequently, models trained only on those outperform chance. Are these correlations picked up by models trained on the full input data? To address this question, we propose a new evaluation method, Counterfactual Attentiveness Test (CAT). CAT uses counterfactuals by replacing part of the input with its counterpart from a different example (subject to some restrictions), expecting an attentive model to change its prediction. Using CAT, we systematically investigate established supervised and in-context learning models on ten datasets spanning four tasks: natural language inference, reading comprehension, paraphrase detection, and visual & language reasoning. CAT reveals that reliance on such correlations is mainly data-dependent. Surprisingly, we find that GPT3 becomes less attentive with an increased number of demonstrations, while its accuracy on the test data improves. Our results demonstrate that augmenting training or demonstration data with counterfactuals is effective in improving models' attentiveness. We show that models' attentiveness measured by CAT reveals different conclusions from solely measuring correlations in data.

Large language models (LLMs) have demonstrated impressive reasoning ability in various language-based tasks. Despite many proposed reasoning methods aimed at enhancing performance in downstream tasks, two fundamental questions persist: Does reasoning genuinely support predictions, and how reliable is the quality of reasoning? In this paper, we propose a framework \textsc{SCORE} to analyze how well LLMs can reason. Specifically, we focus on self-contradictory reasoning, where reasoning does not support the prediction. We find that LLMs often contradict themselves when performing reasoning tasks that involve contextual information and commonsense. The model may miss evidence or use shortcuts, thereby exhibiting self-contradictory behaviors. We also employ the Point-of-View (POV) method, which probes models to generate reasoning from multiple perspectives, as a diagnostic tool for further analysis. We find that though LLMs may appear to perform well in one-perspective settings, they fail to stabilize such behavior in multi-perspectives settings. Even for correct predictions, the reasoning may be messy and incomplete, and LLMs can easily be led astray from good reasoning. \textsc{SCORE}'s results underscore the lack of robustness required for trustworthy reasoning and the urgency for further research to establish best practices for a comprehensive evaluation of reasoning beyond accuracy-based metrics.