The commercialization of large language models (LLMs) has led to the common practice of high-level API-only access to proprietary models. In this work, we show that even with a conservative assumption about the model architecture, it is possible to learn a surprisingly large amount of non-public information about an API-protected LLM from a relatively small number of API queries (e.g., costing under $1,000 for OpenAI's gpt-3.5-turbo). Our findings are centered on one key observation: most modern LLMs suffer from a softmax bottleneck, which restricts the model outputs to a linear subspace of the full output space. We show that this lends itself to a model image or a model signature which unlocks several capabilities with affordable cost: efficiently discovering the LLM's hidden size, obtaining full-vocabulary outputs, detecting and disambiguating different model updates, identifying the source LLM given a single full LLM output, and even estimating the output layer parameters. Our empirical investigations show the effectiveness of our methods, which allow us to estimate the embedding size of OpenAI's gpt-3.5-turbo to be about 4,096. Lastly, we discuss ways that LLM providers can guard against these attacks, as well as how these capabilities can be viewed as a feature (rather than a bug) by allowing for greater transparency and accountability.
Despite their ubiquity in language generation, it remains unknown why truncation sampling heuristics like nucleus sampling are so effective. We provide a theoretical explanation for the effectiveness of the truncation sampling by proving that truncation methods that discard tokens below some probability threshold (the most common type of truncation) can guarantee that all sampled tokens have nonzero true probability. However, thresholds are a coarse heuristic, and necessarily discard some tokens with nonzero true probability as well. In pursuit of a more precise sampling strategy, we show that we can leverage a known source of model errors, the softmax bottleneck, to prove that certain tokens have nonzero true probability, without relying on a threshold. Based on our findings, we develop an experimental truncation strategy and the present pilot studies demonstrating the promise of this type of algorithm. Our evaluations show that our method outperforms its threshold-based counterparts under automatic and human evaluation metrics for low-entropy (i.e., close to greedy) open-ended text generation. Our theoretical findings and pilot experiments provide both insight into why truncation sampling works, and make progress toward more expressive sampling algorithms that better surface the generative capabilities of large language models.
Video summarization remains a huge challenge in computer vision due to the size of the input videos to be summarized. We propose an efficient, language-only video summarizer that achieves competitive accuracy with high data efficiency. Using only textual captions obtained via a zero-shot approach, we train a language transformer model and forego image representations. This method allows us to perform filtration amongst the representative text vectors and condense the sequence. With our approach, we gain explainability with natural language that comes easily for human interpretation and textual summaries of the videos. An ablation study that focuses on modality and data compression shows that leveraging text modality only effectively reduces input data processing while retaining comparable results.
Warning: This paper contains content that may be offensive or upsetting. Understanding the harms and offensiveness of statements requires reasoning about the social and situational context in which statements are made. For example, the utterance "your English is very good" may implicitly signal an insult when uttered by a white man to a non-white colleague, but uttered by an ESL teacher to their student would be interpreted as a genuine compliment. Such contextual factors have been largely ignored by previous approaches to toxic language detection. We introduce COBRA frames, the first context-aware formalism for explaining the intents, reactions, and harms of offensive or biased statements grounded in their social and situational context. We create COBRACORPUS, a dataset of 33k potentially offensive statements paired with machine-generated contexts and free-text explanations of offensiveness, implied biases, speaker intents, and listener reactions. To study the contextual dynamics of offensiveness, we train models to generate COBRA explanations, with and without access to the context. We find that explanations by context-agnostic models are significantly worse than by context-aware ones, especially in situations where the context inverts the statement's offensiveness (29% accuracy drop). Our work highlights the importance and feasibility of contextualized NLP by modeling social factors.
Comparative knowledge (e.g., steel is stronger and heavier than styrofoam) is an essential component of our world knowledge, yet understudied in prior literature. In this paper, we study the task of comparative knowledge acquisition, motivated by the dramatic improvements in the capabilities of extreme-scale language models like GPT-3, which have fueled efforts towards harvesting their knowledge into knowledge bases. However, access to inference API for such models is limited, thereby restricting the scope and the diversity of the knowledge acquisition. We thus ask a seemingly implausible question: whether more accessible, yet considerably smaller and weaker models such as GPT-2, can be utilized to acquire comparative knowledge, such that the resulting quality is on par with their large-scale counterparts? We introduce NeuroComparatives, a novel framework for comparative knowledge distillation using lexically-constrained decoding, followed by stringent filtering of generated knowledge. Our framework acquires comparative knowledge between everyday objects and results in a corpus of 8.7M comparisons over 1.74M entity pairs - 10X larger and 30% more diverse than existing resources. Moreover, human evaluations show that NeuroComparatives outperform existing resources (up to 32% absolute improvement), even including GPT-3, despite using a 100X smaller model. Our results motivate neuro-symbolic manipulation of smaller models as a cost-effective alternative to the currently dominant practice of relying on extreme-scale language models with limited inference access.
Ambiguity is an intrinsic feature of natural language. Managing ambiguity is a key part of human language understanding, allowing us to anticipate misunderstanding as communicators and revise our interpretations as listeners. As language models (LMs) are increasingly employed as dialogue interfaces and writing aids, handling ambiguous language is critical to their success. We characterize ambiguity in a sentence by its effect on entailment relations with another sentence, and collect AmbiEnt, a linguist-annotated benchmark of 1,645 examples with diverse kinds of ambiguity. We design a suite of tests based on AmbiEnt, presenting the first evaluation of pretrained LMs to recognize ambiguity and disentangle possible meanings. We find that the task remains extremely challenging, including for the recent GPT-4, whose generated disambiguations are considered correct only 32% of the time in human evaluation, compared to 90% for disambiguations in our dataset. Finally, to illustrate the value of ambiguity-sensitive tools, we show that a multilabel NLI model can flag political claims in the wild that are misleading due to ambiguity. We encourage the field to rediscover the importance of ambiguity for NLP.
Generative AI has matured to a point where large-scale models can generate text that seems indistinguishable from human-written text and remarkably photorealistic images. Automatically measuring how close the distribution of generated data is to the target real data distribution is a key step in diagnosing existing models and developing better models. We present MAUVE, a family of comparison measures between pairs of distributions such as those encountered in the generative modeling of text or images. These scores are statistical summaries of divergence frontiers capturing two types of errors in generative modeling. We explore four approaches to statistically estimate these scores: vector quantization, non-parametric estimation, classifier-based estimation, and parametric Gaussian approximations. We provide statistical bounds for the vector quantization approach. Empirically, we find that the proposed scores paired with a range of $f$-divergences and statistical estimation methods can quantify the gaps between the distributions of human-written text and those of modern neural language models by correlating with human judgments and identifying known properties of the generated texts. We conclude the paper by demonstrating its applications to other AI domains and discussing practical recommendations.
Pre-trained language models, despite their rapid advancements powered by scale, still fall short of robust commonsense capabilities. And yet, scale appears to be the winning recipe; after all, the largest models seem to have acquired the largest amount of commonsense capabilities. Or is it? In this paper, we investigate the possibility of a seemingly impossible match: can smaller language models with dismal commonsense capabilities (i.e., GPT-2), ever win over models that are orders of magnitude larger and better (i.e., GPT-3), if the smaller models are powered with novel commonsense distillation algorithms? The key intellectual question we ask here is whether it is possible, if at all, to design a learning algorithm that does not benefit from scale, yet leads to a competitive level of commonsense acquisition. In this work, we study the generative models of commonsense knowledge, focusing on the task of generating generics, statements of commonsense facts about everyday concepts, e.g., birds can fly. We introduce a novel commonsense distillation framework, I2D2, that loosely follows the Symbolic Knowledge Distillation of West et al. but breaks the dependence on the extreme-scale models as the teacher model by two innovations: (1) the novel adaptation of NeuroLogic Decoding to enhance the generation quality of the weak, off-the-shelf language models, and (2) self-imitation learning to iteratively learn from the model's own enhanced commonsense acquisition capabilities. Empirical results suggest that scale is not the only way, as novel algorithms can be a promising alternative. Moreover, our study leads to a new corpus of generics, Gen-A-Tomic, that is of the largest and highest quality available to date.
Free-text rationales are a promising step towards explainable AI, yet their evaluation remains an open research problem. While existing metrics have mostly focused on measuring the direct association between the rationale and a given label, we argue that an ideal metric should also be able to focus on the new information uniquely provided in the rationale that is otherwise not provided in the input or the label. We investigate this research problem from an information-theoretic perspective using the conditional V-information. More concretely, we propose a metric called REV (Rationale Evaluation with conditional V-information), that can quantify the new information in a rationale supporting a given label beyond the information already available in the input or the label. Experiments on reasoning tasks across four benchmarks, including few-shot prompting with GPT-3, demonstrate the effectiveness of REV in evaluating different types of rationale-label pairs, compared to existing metrics. Through several quantitative comparisons, we demonstrate the capability of REV in providing more sensitive measurements of new information in free-text rationales with respect to a label. Furthermore, REV is consistent with human judgments on rationale evaluations. Overall, when used alongside traditional performance metrics, REV provides deeper insights into a models' reasoning and prediction processes.
A recurring challenge of crowdsourcing NLP datasets at scale is that human writers often rely on repetitive patterns when crafting examples, leading to a lack of linguistic diversity. We introduce a novel paradigm for dataset creation based on human and machine collaboration, which brings together the generative strength of language models and the evaluative strength of humans. Starting with an existing dataset, MultiNLI, our approach uses dataset cartography to automatically identify examples that demonstrate challenging reasoning patterns, and instructs GPT-3 to compose new examples with similar patterns. Machine generated examples are then automatically filtered, and finally revised and labeled by human crowdworkers to ensure quality. The resulting dataset, WANLI, consists of 108,357 natural language inference (NLI) examples that present unique empirical strengths over existing NLI datasets. Remarkably, training a model on WANLI instead of MNLI (which is 4 times larger) improves performance on seven out-of-domain test sets we consider, including by 11% on HANS and 9% on Adversarial NLI. Moreover, combining MNLI with WANLI is more effective than combining with other augmentation sets that have been introduced. Our results demonstrate the potential of natural language generation techniques to curate NLP datasets of enhanced quality and diversity.