In-context learning$\unicode{x2013}$the ability to configure a model's behavior with different prompts$\unicode{x2013}$has revolutionized the field of natural language processing, alleviating the need for task-specific models and paving the way for generalist models capable of assisting with any query. Computer vision, in contrast, has largely stayed in the former regime: specialized decoders and finetuning protocols are generally required to perform dense tasks such as semantic segmentation and depth estimation. In this work we explore a simple mechanism for in-context learning of such scene understanding tasks: nearest neighbor retrieval from a prompt of annotated features. We propose a new pretraining protocol$\unicode{x2013}$leveraging attention within and across images$\unicode{x2013}$which yields representations particularly useful in this regime. The resulting Hummingbird model, suitably prompted, performs various scene understanding tasks without modification while approaching the performance of specialists that have been finetuned for each task. Moreover, Hummingbird can be configured to perform new tasks much more efficiently than finetuned models, raising the possibility of scene understanding in the interactive assistant regime.
Recent research on dialogue state tracking (DST) focuses on methods that allow few- and zero-shot transfer to new domains or schemas. However, performance gains heavily depend on aggressive data augmentation and fine-tuning of ever larger language model based architectures. In contrast, general purpose language models, trained on large amounts of diverse data, hold the promise of solving any kind of task without task-specific training. We present preliminary experimental results on the ChatGPT research preview, showing that ChatGPT achieves state-of-the-art performance in zero-shot DST. Despite our findings, we argue that properties inherent to general purpose models limit their ability to replace specialized systems. We further theorize that the in-context learning capabilities of such models will likely become powerful tools to support the development of dedicated and dynamic dialogue state trackers.
Text-to-SQL, the process of translating natural language into Structured Query Language (SQL), represents a transformative application of large language models (LLMs), potentially revolutionizing how humans interact with data. This paper introduces the SQL-PaLM framework, a comprehensive solution for understanding and enhancing Text-to-SQL using LLMs, using in the learning regimes of few-shot prompting and instruction fine-tuning. With few-shot prompting, we explore the effectiveness of consistency decoding with execution-based error filtering. With instruction fine-tuning, we delve deep in understanding the critical paradigms that influence the performance of tuned LLMs. In particular, we investigate how performance can be improved through expanded training data coverage and diversity, synthetic data augmentation, and integrating query-specific database content. We propose a test-time selection method to further refine accuracy by integrating SQL outputs from multiple paradigms with execution feedback as guidance. Additionally, we tackle the practical challenge of navigating intricate databases with a significant number of tables and columns, proposing efficient techniques for accurately selecting relevant database elements to enhance Text-to-SQL performance. Our holistic approach yields substantial advancements in Text-to-SQL, as demonstrated on two key public benchmarks, Spider and BIRD. Through comprehensive ablations and error analyses, we shed light on the strengths and weaknesses of our framework, offering valuable insights into Text-to-SQL's future work.
Existing research on Domain Robustness (DR) suffers from disparate setups, limited task variety, and scarce research on recent capabilities such as in-context learning. Furthermore, the common practice of measuring DR might not be fully accurate. Current research focuses on challenge sets and relies solely on the Source Drop (SD): Using the source in-domain performance as a reference point for degradation. However, we argue that the Target Drop (TD), which measures degradation from the target in-domain performance, should be used as a complementary point of view. To address these issues, we first curated a DR benchmark comprised of 7 diverse NLP tasks, which enabled us to measure both the SD and the TD. We then conducted a comprehensive large-scale DR study involving over 14,000 domain shifts across 21 fine-tuned models and few-shot LLMs. We found that both model types suffer from drops upon domain shifts. While fine-tuned models excel in-domain, few-shot LLMs often surpass them cross-domain, showing better robustness. In addition, we found that a large SD can often be explained by shifting to a harder domain rather than by a genuine DR challenge, and this highlights the importance of TD as a complementary metric. We hope our study will shed light on the current DR state of NLP models and promote improved evaluation practices toward more robust models.
Extracting patient information from unstructured text is a critical task in health decision-support and clinical research. Large language models (LLMs) have shown the potential to accelerate clinical curation via few-shot in-context learning, in contrast to supervised learning which requires much more costly human annotations. However, despite drastic advances in modern LLMs such as GPT-4, they still struggle with issues regarding accuracy and interpretability, especially in mission-critical domains such as health. Here, we explore a general mitigation framework using self-verification, which leverages the LLM to provide provenance for its own extraction and check its own outputs. This is made possible by the asymmetry between verification and generation, where the latter is often much easier than the former. Experimental results show that our method consistently improves accuracy for various LLMs in standard clinical information extraction tasks. Additionally, self-verification yields interpretations in the form of a short text span corresponding to each output, which makes it very efficient for human experts to audit the results, paving the way towards trustworthy extraction of clinical information in resource-constrained scenarios. To facilitate future research in this direction, we release our code and prompts.
This paper presents a novel application of large language models in user simulation for task-oriented dialog systems, specifically focusing on an in-context learning approach. By harnessing the power of these models, the proposed approach generates diverse utterances based on user goals and limited dialog examples. Unlike traditional simulators, this method eliminates the need for labor-intensive rule definition or extensive annotated data, making it more efficient and accessible. Additionally, an error analysis of the interaction between the user simulator and dialog system uncovers common mistakes, providing valuable insights into areas that require improvement. Our implementation is available at https://github.com/telepathylabsai/prompt-based-user-simulator.
Intent discovery is the task of inferring latent intents from a set of unlabeled utterances, and is a useful step towards the efficient creation of new conversational agents. We show that recent competitive methods in intent discovery can be outperformed by clustering utterances based on abstractive summaries, i.e., "labels", that retain the core elements while removing non-essential information. We contribute the IDAS approach, which collects a set of descriptive utterance labels by prompting a Large Language Model, starting from a well-chosen seed set of prototypical utterances, to bootstrap an In-Context Learning procedure to generate labels for non-prototypical utterances. The utterances and their resulting noisy labels are then encoded by a frozen pre-trained encoder, and subsequently clustered to recover the latent intents. For the unsupervised task (without any intent labels) IDAS outperforms the state-of-the-art by up to +7.42% in standard cluster metrics for the Banking, StackOverflow, and Transport datasets. For the semi-supervised task (with labels for a subset of intents) IDAS surpasses 2 recent methods on the CLINC benchmark without even using labeled data.
A number of recent benchmarks seek to assess how well models handle natural language negation. However, these benchmarks lack the controlled example paradigms that would allow us to infer whether a model had learned how negation morphemes semantically scope. To fill these analytical gaps, we present the Scoped Negation NLI (ScoNe-NLI) benchmark, which contains contrast sets of six examples with up to two negations where either zero, one, or both negative morphemes affect the NLI label. We use ScoNe-NLI to assess fine-tuning and in-context learning strategies. We find that RoBERTa and DeBERTa models solve ScoNe-NLI after many shot fine-tuning. For in-context learning, we test InstructGPT models and find that most prompt strategies are not successful, including those using step-by-step reasoning. To better understand this result, we extend ScoNe with ScoNe-NLG, a sentence completion test set that embeds negation reasoning in short narratives. Here, InstructGPT is successful, which reveals the model can correctly reason about negation, but struggles to do so on prompt-adapted NLI examples outside of its core pretraining regime.
In this paper, we conduct a comprehensive study of In-Context Learning (ICL) by addressing several open questions: (a) What type of ICL estimator is learned by large language models? (b) What is a proper performance metric for ICL and what is the error rate? (c) How does the transformer architecture enable ICL? To answer these questions, we adopt a Bayesian view and formulate ICL as a problem of predicting the response corresponding to the current covariate, given a number of examples drawn from a latent variable model. To answer (a), we show that, without updating the neural network parameters, ICL implicitly implements the Bayesian model averaging algorithm, which is proven to be approximately parameterized by the attention mechanism. For (b), we analyze the ICL performance from an online learning perspective and establish a $\mathcal{O}(1/T)$ regret bound for perfectly pretrained ICL, where $T$ is the number of examples in the prompt. To answer (c), we show that, in addition to encoding Bayesian model averaging via attention, the transformer architecture also enables a fine-grained statistical analysis of pretraining under realistic assumptions. In particular, we prove that the error of pretrained model is bounded by a sum of an approximation error and a generalization error, where the former decays to zero exponentially as the depth grows, and the latter decays to zero sublinearly with the number of tokens in the pretraining dataset. Our results provide a unified understanding of the transformer and its ICL ability with bounds on ICL regret, approximation, and generalization, which deepens our knowledge of these essential aspects of modern language models.
We introduce Contextual Vision Transformers (ContextViT), a method designed to generate robust image representations for datasets experiencing shifts in latent factors across various groups. Derived from the concept of in-context learning, ContextViT incorporates an additional context token to encapsulate group-specific information. This integration allows the model to adjust the image representation in accordance with the group-specific context. Specifically, for a given input image, ContextViT maps images with identical group membership into this context token, which is appended to the input image tokens. Additionally, we introduce a context inference network to predict such tokens on-the-fly, given a batch of samples from the group. This enables ContextViT to adapt to new testing distributions during inference time. We demonstrate the efficacy of ContextViT across a wide range of applications. In supervised fine-tuning, we show that augmenting pre-trained ViTs with our proposed context conditioning mechanism results in consistent improvements in out-of-distribution generalization on iWildCam and FMoW. We also investigate self-supervised representation learning with ContextViT. Our experiments on the Camelyon17 pathology imaging benchmark and the JUMP-CP microscopy imaging benchmark demonstrate that ContextViT excels in learning stable image featurizations amidst distribution shift, consistently outperforming its ViT counterpart.
We present the training recipe and results of scaling up PaLI-X, a multilingual vision and language model, both in terms of size of the components and the breadth of its training task mixture. Our model achieves new levels of performance on a wide-range of varied and complex tasks, including multiple image-based captioning and question-answering tasks, image-based document understanding and few-shot (in-context) learning, as well as object detection, video question answering, and video captioning. PaLI-X advances the state-of-the-art on most vision-and-language benchmarks considered (25+ of them). Finally, we observe emerging capabilities, such as complex counting and multilingual object detection, tasks that are not explicitly in the training mix.
Various design settings for in-context learning (ICL), such as the choice and order of the in-context examples, can bias a model toward a particular prediction without being reflective of an understanding of the task. While many studies discuss these design choices, there have been few systematic investigations into categorizing them and mitigating their impact. In this work, we define a typology for three types of label biases in ICL for text classification: vanilla-label bias, context-label bias, and domain-label bias (which we conceptualize and detect for the first time). Our analysis demonstrates that prior label bias calibration methods fall short of addressing all three types of biases. Specifically, domain-label bias restricts LLMs to random-level performance on many tasks regardless of the choice of in-context examples. To mitigate the effect of these biases, we propose a simple bias calibration method that estimates a language model's label bias using random in-domain words from the task corpus. After controlling for this estimated bias when making predictions, our novel domain-context calibration significantly improves the ICL performance of GPT-J and GPT-3 on a wide range of tasks. The gain is substantial on tasks with large domain-label bias (up to 37% in Macro-F1). Furthermore, our results generalize to models with different scales, pretraining methods, and manually-designed task instructions, showing the prevalence of label biases in ICL.
Explanations in conventional recommender systems have demonstrated benefits in helping the user understand the rationality of the recommendations and improving the system's efficiency, transparency, and trustworthiness. In the conversational environment, multiple contextualized explanations need to be generated, which poses further challenges for explanations. To better measure explainability in conversational recommender systems (CRS), we propose ten evaluation perspectives based on concepts from conventional recommender systems together with the characteristics of CRS. We assess five existing CRS benchmark datasets using these metrics and observe the necessity of improving the explanation quality of CRS. To achieve this, we conduct manual and automatic approaches to extend these dialogues and construct a new CRS dataset, namely Explainable Recommendation Dialogues (E-ReDial). It includes 756 dialogues with over 2,000 high-quality rewritten explanations. We compare two baseline approaches to perform explanation generation based on E-ReDial. Experimental results suggest that models trained on E-ReDial can significantly improve explainability while introducing knowledge into the models can further improve the performance. GPT-3 in the in-context learning setting can generate more realistic and diverse movie descriptions. In contrast, T5 training on E-ReDial can better generate clear reasons for recommendations based on user preferences. E-ReDial is available at https://github.com/Superbooming/E-ReDial.
Large language models (LLMs) can learn to perform a wide range of natural language tasks from just a handful of in-context examples. However, for generating strings from highly structured languages (e.g., semantic parsing to complex domain-specific languages), it is challenging for the LLM to generalize from just a few exemplars. We propose \emph{grammar prompting}, a simple approach to enable LLMs to use external knowledge and domain-specific constraints, expressed through a grammar in Backus--Naur Form (BNF), during in-context learning. Grammar prompting augments each demonstration example with a specialized grammar that is minimally sufficient for generating the particular output example, where the specialized grammar is a subset of the full DSL grammar. For inference, the LLM first predicts a BNF grammar given a test input, and then generates the output according to the rules of the grammar. Experiments demonstrate that grammar prompting can enable LLMs to perform competitively on a diverse set of DSL generation tasks, including semantic parsing (SMCalFlow, Overnight, GeoQuery), PDDL planning, and SMILES-based molecule generation.
Large language models (LLMs) have a wealth of knowledge that allows them to excel in various Natural Language Processing (NLP) tasks. Current research focuses on enhancing their performance within their existing knowledge. Despite their vast knowledge, LLMs are still limited by the amount of information they can accommodate and comprehend. Therefore, the ability to understand their own limitations on the unknows, referred to as self-knowledge, is of paramount importance. This study aims to evaluate LLMs' self-knowledge by assessing their ability to identify unanswerable or unknowable questions. We introduce an automated methodology to detect uncertainty in the responses of these models, providing a novel measure of their self-knowledge. We further introduce a unique dataset, SelfAware, consisting of unanswerable questions from five diverse categories and their answerable counterparts. Our extensive analysis, involving 20 LLMs including GPT-3, InstructGPT, and LLaMA, discovering an intrinsic capacity for self-knowledge within these models. Moreover, we demonstrate that in-context learning and instruction tuning can further enhance this self-knowledge. Despite this promising insight, our findings also highlight a considerable gap between the capabilities of these models and human proficiency in recognizing the limits of their knowledge.
Using translation memories (TMs) as prompts is a promising approach to in-context learning of machine translation models. In this work, we take a step towards prompting large language models (LLMs) with TMs and making them better translators. We find that the ability of LLMs to ``understand'' prompts is indeed helpful for making better use of TMs. Experiments show that the results of a pre-trained LLM translator can be greatly improved by using high-quality TM-based prompts. These results are even comparable to those of the state-of-the-art NMT systems which have access to large-scale in-domain bilingual data and are well tuned on the downstream tasks.
In this paper, we use Prior-data Fitted Networks (PFNs) as a flexible surrogate for Bayesian Optimization (BO). PFNs are neural processes that are trained to approximate the posterior predictive distribution (PPD) through in-context learning on any prior distribution that can be efficiently sampled from. We describe how this flexibility can be exploited for surrogate modeling in BO. We use PFNs to mimic a naive Gaussian process (GP), an advanced GP, and a Bayesian Neural Network (BNN). In addition, we show how to incorporate further information into the prior, such as allowing hints about the position of optima (user priors), ignoring irrelevant dimensions, and performing non-myopic BO by learning the acquisition function. The flexibility underlying these extensions opens up vast possibilities for using PFNs for BO. We demonstrate the usefulness of PFNs for BO in a large-scale evaluation on artificial GP samples and three different hyperparameter optimization testbeds: HPO-B, Bayesmark, and PD1. We publish code alongside trained models at github.com/automl/PFNs4BO.
Large language models (LLMs) have recently shown great potential for in-context learning, where LLMs learn a new task simply by conditioning on a few input-label pairs (prompts). Despite their potential, our understanding of the factors influencing end-task performance and the robustness of in-context learning remains limited. This paper aims to bridge this knowledge gap by investigating the reliance of LLMs on shortcuts or spurious correlations within prompts. Through comprehensive experiments on classification and extraction tasks, we reveal that LLMs are "lazy learners" that tend to exploit shortcuts in prompts for downstream tasks. Additionally, we uncover a surprising finding that larger models are more likely to utilize shortcuts in prompts during inference. Our findings provide a new perspective on evaluating robustness in in-context learning and pose new challenges for detecting and mitigating the use of shortcuts in prompts.
Analogical reasoning is a fundamental capacity of human cognition that allows us to reason abstractly about novel situations by relating them to past experiences. While it is thought to be essential for robust reasoning in AI systems, conventional approaches require significant training and/or hard-coding of domain knowledge to be applied to benchmark tasks. Inspired by cognitive science research that has found connections between human language and analogy-making, we explore the use of intuitive language-based abstractions to support analogy in AI systems. Specifically, we apply large pre-trained language models (PLMs) to visual Raven's Progressive Matrices (RPM), a common relational reasoning test. By simply encoding the perceptual features of the problem into language form, we find that PLMs exhibit a striking capacity for zero-shot relational reasoning, exceeding human performance and nearing supervised vision-based methods. We explore different encodings that vary the level of abstraction over task features, finding that higher-level abstractions further strengthen PLMs' analogical reasoning. Our detailed analysis reveals insights on the role of model complexity, in-context learning, and prior knowledge in solving RPM tasks.
Few-shot fine-tuning and in-context learning are two alternative strategies for task adaptation of pre-trained language models. Recently, in-context learning has gained popularity over fine-tuning due to its simplicity and improved out-of-domain generalization, and because extensive evidence shows that fine-tuned models pick up on spurious correlations. Unfortunately, previous comparisons of the two approaches were done using models of different sizes. This raises the question of whether the observed weaker out-of-domain generalization of fine-tuned models is an inherent property of fine-tuning or a limitation of the experimental setup. In this paper, we compare the generalization of few-shot fine-tuning and in-context learning to challenge datasets, while controlling for the models used, the number of examples, and the number of parameters, ranging from 125M to 30B. Our results show that fine-tuned language models can in fact generalize well out-of-domain. We find that both approaches generalize similarly; they exhibit large variation and depend on properties such as model size and the number of examples, highlighting that robust task adaptation remains a challenge.