Derek Chen

Aarohi Srivastava, Abhinav Rastogi, Abhishek Rao et al. · allen-ai, amazon-science

Language models demonstrate both quantitative improvement and new qualitative capabilities with increasing scale. Despite their potentially transformative impact, these new capabilities are as yet poorly characterized. In order to inform future research, prepare for disruptive new model capabilities, and ameliorate socially harmful effects, it is vital that we understand the present and near-future capabilities and limitations of language models. To address this challenge, we introduce the Beyond the Imitation Game benchmark (BIG-bench). BIG-bench currently consists of 204 tasks, contributed by 450 authors across 132 institutions. Task topics are diverse, drawing problems from linguistics, childhood development, math, common-sense reasoning, biology, physics, social bias, software development, and beyond. BIG-bench focuses on tasks that are believed to be beyond the capabilities of current language models. We evaluate the behavior of OpenAI's GPT models, Google-internal dense transformer architectures, and Switch-style sparse transformers on BIG-bench, across model sizes spanning millions to hundreds of billions of parameters. In addition, a team of human expert raters performed all tasks in order to provide a strong baseline. Findings include: model performance and calibration both improve with scale, but are poor in absolute terms (and when compared with rater performance); performance is remarkably similar across model classes, though with benefits from sparsity; tasks that improve gradually and predictably commonly involve a large knowledge or memorization component, whereas tasks that exhibit "breakthrough" behavior at a critical scale often involve multiple steps or components, or brittle metrics; social bias typically increases with scale in settings with ambiguous context, but this can be improved with prompting.

Josh Achiam, Steven Adler, Sandhini Agarwal et al. · berkeley, deepmind

We report the development of GPT-4, a large-scale, multimodal model which can accept image and text inputs and produce text outputs. While less capable than humans in many real-world scenarios, GPT-4 exhibits human-level performance on various professional and academic benchmarks, including passing a simulated bar exam with a score around the top 10% of test takers. GPT-4 is a Transformer-based model pre-trained to predict the next token in a document. The post-training alignment process results in improved performance on measures of factuality and adherence to desired behavior. A core component of this project was developing infrastructure and optimization methods that behave predictably across a wide range of scales. This allowed us to accurately predict some aspects of GPT-4's performance based on models trained with no more than 1/1,000th the compute of GPT-4.

Aaditya Singh, Adam Fry, Adam Perelman et al. · berkeley, mila

This is the system card published alongside the OpenAI GPT-5 launch, August 2025. GPT-5 is a unified system with a smart and fast model that answers most questions, a deeper reasoning model for harder problems, and a real-time router that quickly decides which model to use based on conversation type, complexity, tool needs, and explicit intent (for example, if you say 'think hard about this' in the prompt). The router is continuously trained on real signals, including when users switch models, preference rates for responses, and measured correctness, improving over time. Once usage limits are reached, a mini version of each model handles remaining queries. This system card focuses primarily on gpt-5-thinking and gpt-5-main, while evaluations for other models are available in the appendix. The GPT-5 system not only outperforms previous models on benchmarks and answers questions more quickly, but -- more importantly -- is more useful for real-world queries. We've made significant advances in reducing hallucinations, improving instruction following, and minimizing sycophancy, and have leveled up GPT-5's performance in three of ChatGPT's most common uses: writing, coding, and health. All of the GPT-5 models additionally feature safe-completions, our latest approach to safety training to prevent disallowed content. Similarly to ChatGPT agent, we have decided to treat gpt-5-thinking as High capability in the Biological and Chemical domain under our Preparedness Framework, activating the associated safeguards. While we do not have definitive evidence that this model could meaningfully help a novice to create severe biological harm -- our defined threshold for High capability -- we have chosen to take a precautionary approach.

Derek Chen, Celine Lee, Yunan Lu et al.

Large language models (LLMs) effectively generate fluent text when the target output follows natural language patterns. However, structured prediction tasks confine the output format to a limited ontology, causing even very large models to struggle since they were never trained with such restrictions in mind. The difficulty of using LLMs for direct prediction is exacerbated in few-shot learning scenarios, which commonly arise due to domain shift and resource limitations. We flip the problem on its head by leveraging the LLM as a tool for data augmentation rather than direct prediction. Our proposed Mixture of Soft Prompts (MSP) serves as a parameter-efficient procedure for generating data in a controlled manner. Denoising mechanisms are further applied to improve the quality of synthesized data. Automatic metrics show our method is capable of producing diverse and natural text, while preserving label semantics. Moreover, MSP achieves state-of-the-art results on three benchmarks when compared against strong baselines. Our method offers an alternate data-centric approach for applying LLMs to complex prediction tasks.

Derek Chen, Kun Qian, Zhou Yu

Prompt-based methods with large pre-trained language models (PLMs) have shown impressive unaided performance across many NLP tasks. These models improve even further with the addition of a few labeled in-context exemplars to guide output generation. However, for more complex tasks such as dialogue state tracking (DST), designing prompts that reliably convey the desired intent is nontrivial, leading to unstable results. Furthermore, building in-context exemplars for dialogue tasks is difficult because conversational contexts are long while model input lengths are relatively short. To overcome these issues we first adapt a meta-learning scheme to the dialogue domain which stabilizes the ability of the model to perform well under various prompts. We additionally design a novel training method to improve upon vanilla retrieval mechanisms to find ideal in-context examples. Finally, we introduce a saliency model to limit dialogue text length, allowing us to include more exemplars per query. In effect, we are able to achieve highly competitive results for few-shot DST on MultiWOZ.

Justin T. Chiu, Wenting Zhao, Derek Chen et al.

Large language models (LLMs) excel at processing and generating both text and code. However, LLMs have had limited applicability in grounded task-oriented dialogue as they are difficult to steer toward task objectives and fail to handle novel grounding. We present a modular and interpretable grounded dialogue system that addresses these shortcomings by composing LLMs with a symbolic planner and grounded code execution. Our system consists of a reader and planner: the reader leverages an LLM to convert partner utterances into executable code, calling functions that perform grounding. The translated code's output is stored to track dialogue state, while a symbolic planner determines the next appropriate response. We evaluate our system's performance on the demanding OneCommon dialogue task, involving collaborative reference resolution on abstract images of scattered dots. Our system substantially outperforms the previous state-of-the-art, including improving task success in human evaluations from 56% to 69% in the most challenging setting.

Derek Chen, Zhou Yu

Training dialogue systems often entails dealing with noisy training examples and unexpected user inputs. Despite their prevalence, there currently lacks an accurate survey of dialogue noise, nor is there a clear sense of the impact of each noise type on task performance. This paper addresses this gap by first constructing a taxonomy of noise encountered by dialogue systems. In addition, we run a series of experiments to show how different models behave when subjected to varying levels of noise and types of noise. Our results reveal that models are quite robust to label errors commonly tackled by existing denoising algorithms, but that performance suffers from dialogue-specific noise. Driven by these observations, we design a data cleaning algorithm specialized for conversational settings and apply it as a proof-of-concept for targeted dialogue denoising.

Derek Chen, Claire Yin

Training accurate intent classifiers requires labeled data, which can be costly to obtain. Data augmentation methods may ameliorate this issue, but the quality of the generated data varies significantly across techniques. We study the process of systematically producing pseudo-labeled data given a small seed set using a wide variety of data augmentation techniques, including mixing methods together. We find that while certain methods dramatically improve qualitative and quantitative performance, other methods have minimal or even negative impact. We also analyze key considerations when implementing data augmentation methods in production.

Qingyang Wu, Deema Alnuhait, Derek Chen et al.

Traditional end-to-end task-oriented dialogue systems have been built with a modularized design. However, such design often causes misalignment between the agent response and external knowledge, due to inadequate representation of information. Furthermore, its evaluation metrics emphasize assessing the agent's pre-lexicalization response, neglecting the quality of the completed response. In this work, we propose a novel paradigm that uses a textual interface to align external knowledge and eliminate redundant processes. We demonstrate our paradigm in practice through MultiWOZ-Remake, including an interactive textual interface built for the MultiWOZ database and a correspondingly re-processed dataset. We train an end-to-end dialogue system to evaluate this new dataset. The experimental results show that our approach generates more natural final responses and achieves a greater task success rate compared to the previous models.

Qingyang Wu, Song Feng, Derek Chen et al.

Collecting data for training dialog systems can be extremely expensive due to the involvement of human participants and need for extensive annotation. Especially in document-grounded dialog systems, human experts need to carefully read the unstructured documents to answer the users' questions. As a result, existing document-grounded dialog datasets are relatively small-scale and obstruct the effective training of dialogue systems. In this paper, we propose an automatic data augmentation technique grounded on documents through a generative dialogue model. The dialogue model consists of a user bot and agent bot that can synthesize diverse dialogues given an input document, which are then used to train a downstream model. When supplementing the original dataset, our method achieves significant improvement over traditional data augmentation methods. We also achieve great performance in the low-resource setting.

Derek Chen, Zhou Yu, Samuel R. Bowman

Crowdsourcing platforms are often used to collect datasets for training machine learning models, despite higher levels of inaccurate labeling compared to expert labeling. There are two common strategies to manage the impact of such noise. The first involves aggregating redundant annotations, but comes at the expense of labeling substantially fewer examples. Secondly, prior works have also considered using the entire annotation budget to label as many examples as possible and subsequently apply denoising algorithms to implicitly clean the dataset. We find a middle ground and propose an approach which reserves a fraction of annotations to explicitly clean up highly probable error samples to optimize the annotation process. In particular, we allocate a large portion of the labeling budget to form an initial dataset used to train a model. This model is then used to identify specific examples that appear most likely to be incorrect, which we spend the remaining budget to relabel. Experiments across three model variations and four natural language processing tasks show our approach outperforms or matches both label aggregation and advanced denoising methods designed to handle noisy labels when allocated the same finite annotation budget.

Derek Chen, Zhou Yu

Practical dialogue systems require robust methods of detecting out-of-scope (OOS) utterances to avoid conversational breakdowns and related failure modes. Directly training a model with labeled OOS examples yields reasonable performance, but obtaining such data is a resource-intensive process. To tackle this limited-data problem, previous methods focus on better modeling the distribution of in-scope (INS) examples. We introduce GOLD as an orthogonal technique that augments existing data to train better OOS detectors operating in low-data regimes. GOLD generates pseudo-labeled candidates using samples from an auxiliary dataset and keeps only the most beneficial candidates for training through a novel filtering mechanism. In experiments across three target benchmarks, the top GOLD model outperforms all existing methods on all key metrics, achieving relative gains of 52.4%, 48.9% and 50.3% against median baseline performance. We also analyze the unique properties of OOS data to identify key factors for optimally applying our proposed method.

Derek Chen, Howard Chen, Yi Yang et al.

Existing goal-oriented dialogue datasets focus mainly on identifying slots and values. However, customer support interactions in reality often involve agents following multi-step procedures derived from explicitly-defined company policies as well. To study customer service dialogue systems in more realistic settings, we introduce the Action-Based Conversations Dataset (ABCD), a fully-labeled dataset with over 10K human-to-human dialogues containing 55 distinct user intents requiring unique sequences of actions constrained by policies to achieve task success. We propose two additional dialog tasks, Action State Tracking and Cascading Dialogue Success, and establish a series of baselines involving large-scale, pre-trained language models on this dataset. Empirical results demonstrate that while more sophisticated networks outperform simpler models, a considerable gap (50.8% absolute accuracy) still exists to reach human-level performance on ABCD.

Maarten Sap, Hannah Rashkin, Derek Chen et al.

We introduce Social IQa, the first largescale benchmark for commonsense reasoning about social situations. Social IQa contains 38,000 multiple choice questions for probing emotional and social intelligence in a variety of everyday situations (e.g., Q: "Jordan wanted to tell Tracy a secret, so Jordan leaned towards Tracy. Why did Jordan do this?" A: "Make sure no one else could hear"). Through crowdsourcing, we collect commonsense questions along with correct and incorrect answers about social interactions, using a new framework that mitigates stylistic artifacts in incorrect answers by asking workers to provide the right answer to a different but related question. Empirical results show that our benchmark is challenging for existing question-answering models based on pretrained language models, compared to human performance (>20% gap). Notably, we further establish Social IQa as a resource for transfer learning of commonsense knowledge, achieving state-of-the-art performance on multiple commonsense reasoning tasks (Winograd Schemas, COPA).

He He, Derek Chen, Anusha Balakrishnan et al.

We consider negotiation settings in which two agents use natural language to bargain on goods. Agents need to decide on both high-level strategy (e.g., proposing \$50) and the execution of that strategy (e.g., generating "The bike is brand new. Selling for just \$50."). Recent work on negotiation trains neural models, but their end-to-end nature makes it hard to control their strategy, and reinforcement learning tends to lead to degenerate solutions. In this paper, we propose a modular approach based on coarse di- alogue acts (e.g., propose(price=50)) that decouples strategy and generation. We show that we can flexibly set the strategy using supervised learning, reinforcement learning, or domain-specific knowledge without degeneracy, while our retrieval-based generation can maintain context-awareness and produce diverse utterances. We test our approach on the recently proposed DEALORNODEAL game, and we also collect a richer dataset based on real items on Craigslist. Human evaluation shows that our systems achieve higher task success rate and more human-like negotiation behavior than previous approaches.