Transformers are remarkably good at in-context learning (ICL) -- learning from demonstrations without parameter updates -- but how they perform ICL remains a mystery. Recent work suggests that Transformers may learn in-context by internally running Gradient Descent, a first-order optimization method. In this paper, we instead demonstrate that Transformers learn to implement higher-order optimization methods to perform ICL. Focusing on in-context linear regression, we show that Transformers learn to implement an algorithm very similar to Iterative Newton's Method, a higher-order optimization method, rather than Gradient Descent. Empirically, we show that predictions from successive Transformer layers closely match different iterations of Newton's Method linearly, with each middle layer roughly computing 3 iterations. In contrast, exponentially more Gradient Descent steps are needed to match an additional Transformers layer; this suggests that Transformers have an comparable rate of convergence with high-order methods such as Iterative Newton, which are exponentially faster than Gradient Descent. We also show that Transformers can learn in-context on ill-conditioned data, a setting where Gradient Descent struggles but Iterative Newton succeeds. Finally, we show theoretical results which support our empirical findings and have a close correspondence with them: we prove that Transformers can implement $k$ iterations of Newton's method with $\mathcal{O}(k)$ layers.

Do question answering (QA) modeling improvements (e.g., choice of architecture and training procedure) hold consistently across the diverse landscape of QA benchmarks? To study this question, we introduce the notion of concurrence -- two benchmarks have high concurrence on a set of modeling approaches if they rank the modeling approaches similarly. We measure the concurrence between 32 QA benchmarks on a set of 20 diverse modeling approaches and find that human-constructed benchmarks have high concurrence amongst themselves, even if their passage and question distributions are very different. Surprisingly, even downsampled human-constructed benchmarks (i.e., collecting less data) and programmatically-generated benchmarks (e.g., cloze-formatted examples) have high concurrence with human-constructed benchmarks. These results indicate that, despite years of intense community focus on a small number of benchmarks, the modeling improvements studied hold broadly.

Recent results in image classification and extractive question answering have observed that pre-trained models trained on less in-distribution data have better out-of-distribution performance. However, it is unclear how broadly these trends hold. We conduct a large empirical study across three tasks, three broadly-applicable modeling interventions (increasing model size, using a different adaptation method, and pre-training on more data), and 14 diverse datasets to investigate the relationship between sample efficiency (amount of data needed to reach a given ID accuracy) and robustness (how models fare on OOD evaluation). We find that higher sample efficiency is only correlated with better average OOD robustness on some modeling interventions and tasks, but not others. On individual datasets, models with lower sample efficiency can even be more robust. These results suggest that general-purpose methods for improving sample efficiency are unlikely to yield universal OOD robustness improvements, since such improvements are highly dataset- and task-dependent. Even in an era of large, multi-purpose pretrained models, task-specific decisions may often be necessary for OOD generalization.

In many task settings, text classification models are likely to encounter examples from novel classes on which they cannot predict correctly. Selective prediction, in which models abstain on low-confidence examples, provides a possible solution, but existing models are often overly confident on unseen classes. To remedy this overconfidence, we introduce Contrastive Novelty-Augmented Learning (CoNAL), a two-step method that generates OOD examples representative of novel classes, then trains to decrease confidence on them. First, we generate OOD examples by prompting a large language model twice: we prompt it to enumerate relevant novel classes, then generate examples from each novel class matching the task format. Second, we train a classifier with a novel contrastive objective that encourages lower confidence on generated OOD examples than training examples. When trained with CoNAL, classifiers improve in their ability to detect and abstain on novel class examples over prior methods by an average of 2.3% in terms of accuracy under the accuracy-coverage curve (AUAC) and 5.5% AUROC across 4 NLP datasets, with no cost to in-distribution accuracy.

In-context learning (ICL) enables large language models (LLMs) to perform new tasks by prompting them with a sequence of training examples. However, it is known that ICL is very sensitive to the choice of training examples: randomly sampling examples from a training set leads to high variance in performance. In this paper, we show that carefully curating a subset of training data greatly stabilizes ICL performance without any other changes to the ICL algorithm (e.g., prompt retrieval or calibration). We introduce two methods to choose training subsets -- both score training examples individually, then select the highest-scoring ones. CondAcc scores a training example by its average dev-set ICL accuracy when combined with random training examples, while Datamodels learns linear regressors that estimate how the presence of each training example influences LLM outputs. Across five tasks and two LLMs, sampling from stable subsets selected by CondAcc and Datamodels improves average accuracy over sampling from the entire training set by 7.7% and 6.3%, respectively. Surprisingly, the stable subset examples are not especially diverse in content or low in perplexity, in contrast with other work suggesting that diversity and perplexity are important when prompting LLMs.

In order to reliably process natural language, NLP systems must generalize to the long tail of rare utterances. We propose a method to create challenging benchmarks that require generalizing to the tail of the distribution by re-splitting existing datasets. We create 'Likelihood Splits' where examples that are assigned lower likelihood by a pre-trained language model (LM) are placed in the test set, and more likely examples are in the training set. This simple approach can be customized to construct meaningful train-test splits for a wide range of tasks. Likelihood Splits surface more challenges than random splits: relative error rates of state-of-the-art models increase by 59% for semantic parsing on Spider, 93% for natural language inference on SNLI, and 33% for yes/no question answering on BoolQ, on our splits compared with the corresponding random splits. Moreover, Likelihood Splits create fairer benchmarks than adversarial filtering; when the LM used to create the splits is also employed as the task model, our splits do not unfairly penalize the LM.

We introduce Dynaboard, an evaluation-as-a-service framework for hosting benchmarks and conducting holistic model comparison, integrated with the Dynabench platform. Our platform evaluates NLP models directly instead of relying on self-reported metrics or predictions on a single dataset. Under this paradigm, models are submitted to be evaluated in the cloud, circumventing the issues of reproducibility, accessibility, and backwards compatibility that often hinder benchmarking in NLP. This allows users to interact with uploaded models in real time to assess their quality, and permits the collection of additional metrics such as memory use, throughput, and robustness, which -- despite their importance to practitioners -- have traditionally been absent from leaderboards. On each task, models are ranked according to the Dynascore, a novel utility-based aggregation of these statistics, which users can customize to better reflect their preferences, placing more/less weight on a particular axis of evaluation or dataset. As state-of-the-art NLP models push the limits of traditional benchmarks, Dynaboard offers a standardized solution for a more diverse and comprehensive evaluation of model quality.

We introduce Dynabench, an open-source platform for dynamic dataset creation and model benchmarking. Dynabench runs in a web browser and supports human-and-model-in-the-loop dataset creation: annotators seek to create examples that a target model will misclassify, but that another person will not. In this paper, we argue that Dynabench addresses a critical need in our community: contemporary models quickly achieve outstanding performance on benchmark tasks but nonetheless fail on simple challenge examples and falter in real-world scenarios. With Dynabench, dataset creation, model development, and model assessment can directly inform each other, leading to more robust and informative benchmarks. We report on four initial NLP tasks, illustrating these concepts and highlighting the promise of the platform, and address potential objections to dynamic benchmarking as a new standard for the field.

Despite its importance to experimental design, statistical power (the probability that, given a real effect, an experiment will reject the null hypothesis) has largely been ignored by the NLP community. Underpowered experiments make it more difficult to discern the difference between statistical noise and meaningful model improvements, and increase the chances of exaggerated findings. By meta-analyzing a set of existing NLP papers and datasets, we characterize typical power for a variety of settings and conclude that underpowered experiments are common in the NLP literature. In particular, for several tasks in the popular GLUE benchmark, small test sets mean that most attempted comparisons to state of the art models will not be adequately powered. Similarly, based on reasonable assumptions, we find that the most typical experimental design for human rating studies will be underpowered to detect small model differences, of the sort that are frequently studied. For machine translation, we find that typical test sets of 2000 sentences have approximately 75% power to detect differences of 1 BLEU point. To improve the situation going forward, we give an overview of best practices for power analysis in NLP and release a series of notebooks to assist with future power analyses.