arXiv subjects

David Sontag

Publications and source records attributed to David Sontag.

At least 19 recordsLinked to original sources

arXiv2026

LLMs can construct powerful representations and streamline sample-efficient supervised learning

As real-world datasets become more complex and heterogeneous, supervised learning is often bottlenecked by input representation design. Modeling multimodal data, such as time-series, free text, and structured records, often requires non-trivial domain expertise. We propose an agentic pipeline to streamline this process. First, an LLM analyzes a small but diverse subset of text-serialized input examples in-context to synthesize a global rubric, which acts as a programmatic specification for extracting and organizing evidence. This rubric is then used to transform naive text-serializations of inputs into a more standardized format for downstream models. We also describe local rubrics, which are task-conditioned interpretive summaries generated by an LLM. Across 15 clinical tasks from the EHRSHOT benchmark, our rubric approaches significantly outperform count-feature models, naive LLM baselines, and a clinical foundation model pretrained on orders of magnitude more data. Beyond performance, rubrics offer operational advantages such as being easy to audit, cost-effectiveness at scale, and facilitating tabular representations.

cs.AI↗

arXiv2026

Evaluating Physician-AI Interaction for Cancer Management: Paving the Path towards Precision Oncology

As machine learning (ML)-based decision support tools proliferate in clinical practice, understanding how clinicians integrate personalized ML predictions alongside randomized controlled trial (RCT) evidence is critical. We designed a web-based clinical decision support system (CDSS) presenting survival and adverse event data from a simulated RCT and ML model across 12 synthetic multiple myeloma scenarios. In a within- subjects study with 32 physicians, we evaluated how clinicians synthesize competing evidence sources to make treatment decisions. When ML and RCT outputs were concordant, physicians reported greater confidence than with RCT data alone. When results were discordant, most physicians shifted toward the ML-supported treatment, often before reviewing any information about model training or validation, suggesting a tendency toward automation bias rather than algorithm avoidance. Despite reporting higher perceived reliability after viewing model quality disclosures, physicians were largely unable to describe the validation procedures they had reviewed. Taken together, these findings reveal that clinicians may over-rely on ML recommendations even when equipped with tools designed to support critical appraisal. We discuss implications for CDSS design, clinician training, and the institutional safeguards needed before ML-based systems are deployed in high-stakes oncology settings.

cs.HC↗

arXiv2026

DR Tulu: Reinforcement Learning with Evolving Rubrics for Deep Research

Deep research agents perform multi-step research to produce long-form, well-attributed answers. However, most open deep research agents are trained on easily verifiable short-form QA tasks via reinforcement learning with verifiable rewards, which does not extend to realistic long-form tasks. We address this with Reinforcement Learning with Evolving Rubrics (RLER), where rubrics are constructed and maintained to co-evolve with the policy model during training. This allows the rubrics to incorporate newly explored information from search and contrasting model responses, enabling better fact checking and more discriminative on-policy feedback. Using RLER, we develop Deep Research Tulu (DR Tulu-8B), the first fully open model that is directly trained for open-ended, long-form deep research. Across four long-form deep research benchmarks in science, healthcare, and general domains, DR Tulu substantially outperforms existing open deep research agents (by 15.6% over Tongyi DR on average) and matches or exceeds proprietary deep research agents (by 0.7% over OpenAI DR on average), while being significantly smaller and cheaper per query (1000x cheaper than OpenAI DR per query).

cs.CL↗

arXiv2025

Completion $\neq$ Collaboration: Scaling Collaborative Effort with Agents

Current evaluations of agents remain centered around one-shot task completion, failing to account for the inherently iterative and collaborative nature of many real-world problems, where human goals are often underspecified and evolve. We argue for a shift from building and assessing task completion agents to developing collaborative agents, assessed not only by the quality of their final outputs but by how well they engage with and enhance human effort throughout the problem-solving process. To support this shift, we introduce collaborative effort scaling, a framework that captures how an agent's utility grows with increasing user involvement. Through case studies and simulated evaluations, we show that state-of-the-art agents often underperform in multi-turn, real-world scenarios, revealing a missing ingredient in agent design: the ability to sustain engagement and scaffold user understanding. Collaborative effort scaling offers a lens for diagnosing agent behavior and guiding development toward more effective interactions.

cs.CL↗

arXiv2025

Uncovering Bias Mechanisms in Observational Studies

Observational studies are a key resource for causal inference but are often affected by systematic biases. Prior work has focused mainly on detecting these biases, via sensitivity analyses and comparisons with randomized controlled trials, or mitigating them through debiasing techniques. However, there remains a lack of methodology for uncovering the underlying mechanisms driving these biases, e.g., whether due to hidden confounding or selection of participants. In this work, we show that the relationship between bias magnitude and the predictive performance of nuisance function estimators (in the observational study) can help distinguish among common sources of causal bias. We validate our methodology through extensive synthetic experiments and a real-world case study, demonstrating its effectiveness in revealing the mechanisms behind observed biases. Our framework offers a new lens for understanding and characterizing bias in observational studies, with practical implications for improving causal inference.

stat.ME↗

arXiv2025

Diagnosing our datasets: How does my language model learn clinical information?

Large language models (LLMs) have performed well across various clinical natural language processing tasks, despite not being directly trained on electronic health record (EHR) data. In this work, we examine how popular open-source LLMs learn clinical information from large mined corpora through two crucial but understudied lenses: (1) their interpretation of clinical jargon, a foundational ability for understanding real-world clinical notes, and (2) their responses to unsupported medical claims. For both use cases, we investigate the frequency of relevant clinical information in their corresponding pretraining corpora, the relationship between pretraining data composition and model outputs, and the sources underlying this data. To isolate clinical jargon understanding, we evaluate LLMs on a new dataset MedLingo. Unsurprisingly, we find that the frequency of clinical jargon mentions across major pretraining corpora correlates with model performance. However, jargon frequently appearing in clinical notes often rarely appears in pretraining corpora, revealing a mismatch between available data and real-world usage. Similarly, we find that a non-negligible portion of documents support disputed claims that can then be parroted by models. Finally, we classified and analyzed the types of online sources in which clinical jargon and unsupported medical claims appear, with implications for future dataset composition.

cs.CL↗

arXiv2025

CodingGenie: A Proactive LLM-Powered Programming Assistant

While developers increasingly adopt tools powered by large language models (LLMs) in day-to-day workflows, these tools still require explicit user invocation. To seamlessly integrate LLM capabilities to a developer's workflow, we introduce CodingGenie, a proactive assistant integrated into the code editor. CodingGenie autonomously provides suggestions, ranging from bug fixing to unit testing, based on the current code context and allows users to customize suggestions by providing a task description and selecting what suggestions are shown. We demonstrate multiple use cases to show how proactive suggestions from CodingGenie can improve developer experience, and also analyze the cost of adding proactivity. We believe this open-source tool will enable further research into proactive assistants. CodingGenie is open-sourced at https://github.com/sebzhao/CodingGenie/ and video demos are available at https://sebzhao.github.io/CodingGenie/.

cs.HC↗

arXiv2025

Need Help? Designing Proactive AI Assistants for Programming

While current chat-based AI assistants primarily operate reactively, responding only when prompted by users, there is significant potential for these systems to proactively assist in tasks without explicit invocation, enabling a mixed-initiative interaction. This work explores the design and implementation of proactive AI assistants powered by large language models. We first outline the key design considerations for building effective proactive assistants. As a case study, we propose a proactive chat-based programming assistant that automatically provides suggestions and facilitates their integration into the programmer's code. The programming context provides a shared workspace enabling the assistant to offer more relevant suggestions. We conducted a randomized experimental study examining the impact of various design elements of the proactive assistant on programmer productivity and user experience. Our findings reveal significant benefits of incorporating proactive chat assistants into coding environments and uncover important nuances that influence their usage and effectiveness.

cs.HC↗

arXiv2025

Large Language Models are Powerful Electronic Health Record Encoders

Electronic Health Records (EHRs) offer considerable potential for clinical prediction, but their complexity and heterogeneity challenge traditional machine learning. Domain-specific EHR foundation models trained on unlabeled EHR data have shown improved predictive accuracy and generalization. However, their development is constrained by limited data access and site-specific vocabularies. We convert EHR data into plain text by replacing medical codes with natural-language descriptions, enabling general-purpose Large Language Models (LLMs) to produce high-dimensional embeddings for downstream prediction tasks without access to private medical training data. LLM-based embeddings perform on par with a specialized EHR foundation model, CLMBR-T-Base, across 15 clinical tasks from the EHRSHOT benchmark. In an external validation using the UK Biobank, an LLM-based model shows statistically significant improvements for some tasks, which we attribute to higher vocabulary coverage and slightly better generalization. Overall, we reveal a trade-off between the computational efficiency of specialized EHR models and the portability and data independence of LLM-based embeddings.

cs.LG↗

arXiv2024

Probably approximately correct high-dimensional causal effect estimation given a valid adjustment set

Accurate estimates of causal effects play a key role in decision-making across applications such as healthcare, economics, and operations. In the absence of randomized experiments, a common approach to estimating causal effects uses \textit{covariate adjustment}. In this paper, we study covariate adjustment for discrete distributions from the PAC learning perspective, assuming knowledge of a valid adjustment set $\bZ$, which might be high-dimensional. Our first main result PAC-bounds the estimation error of covariate adjustment by a term that is exponential in the size of the adjustment set; it is known that such a dependency is unavoidable even if one only aims to minimize the mean squared error. Motivated by this result, we introduce the notion of an \emph{$\eps$-Markov blanket}, give bounds on the misspecification error of using such a set for covariate adjustment, and provide an algorithm for $\eps$-Markov blanket discovery; our second main result upper bounds the sample complexity of this algorithm. Furthermore, we provide a misspecification error bound and a constraint-based algorithm that allow us to go beyond $\eps$-Markov blankets to even smaller adjustment sets. Our third main result upper bounds the sample complexity of this algorithm, and our final result combines the first three into an overall PAC bound. Altogether, our results highlight that one does not need to perfectly recover causal structure in order to ensure accurate estimates of causal effects.

math.ST↗

arXiv2024

Med-Real2Sim: Non-Invasive Medical Digital Twins using Physics-Informed Self-Supervised Learning

A digital twin is a virtual replica of a real-world physical phenomena that uses mathematical modeling to characterize and simulate its defining features. By constructing digital twins for disease processes, we can perform in-silico simulations that mimic patients' health conditions and counterfactual outcomes under hypothetical interventions in a virtual setting. This eliminates the need for invasive procedures or uncertain treatment decisions. In this paper, we propose a method to identify digital twin model parameters using only noninvasive patient health data. We approach the digital twin modeling as a composite inverse problem, and observe that its structure resembles pretraining and finetuning in self-supervised learning (SSL). Leveraging this, we introduce a physics-informed SSL algorithm that initially pretrains a neural network on the pretext task of learning a differentiable simulator of a physiological process. Subsequently, the model is trained to reconstruct physiological measurements from noninvasive modalities while being constrained by the physical equations learned in pretraining. We apply our method to identify digital twins of cardiac hemodynamics using noninvasive echocardiogram videos, and demonstrate its utility in unsupervised disease detection and in-silico clinical trials.

cs.LG↗

arXiv2024

Impact of Large Language Model Assistance on Patients Reading Clinical Notes: A Mixed-Methods Study

Large language models (LLMs) have immense potential to make information more accessible, particularly in medicine, where complex medical jargon can hinder patient comprehension of clinical notes. We developed a patient-facing tool using LLMs to make clinical notes more readable by simplifying, extracting information from, and adding context to the notes. We piloted the tool with clinical notes donated by patients with a history of breast cancer and synthetic notes from a clinician. Participants (N=200, healthy, female-identifying patients) were randomly assigned three clinical notes in our tool with varying levels of augmentations and answered quantitative and qualitative questions evaluating their understanding of follow-up actions. Augmentations significantly increased their quantitative understanding scores. In-depth interviews were conducted with participants (N=7, patients with a history of breast cancer), revealing both positive sentiments about the augmentations and concerns about AI. We also performed a qualitative clinician-driven analysis of the model's error modes.

cs.HC↗

arXiv2024

The RealHumanEval: Evaluating Large Language Models' Abilities to Support Programmers

Evaluation of large language models for code has primarily relied on static benchmarks, including HumanEval (Chen et al., 2021), or more recently using human preferences of LLM responses. As LLMs are increasingly used as programmer assistants, we study whether gains on existing benchmarks or more preferred LLM responses translate to programmer productivity when coding with LLMs, including time spent coding. We introduce RealHumanEval, a web interface to measure the ability of LLMs to assist programmers, through either autocomplete or chat support. We conducted a user study (N=243) using RealHumanEval in which users interacted with seven LLMs of varying base model performance. Despite static benchmarks not incorporating humans-in-the-loop, we find that improvements in benchmark performance lead to increased programmer productivity; however gaps in benchmark versus human performance are not proportional -- a trend that holds across both forms of LLM support. In contrast, we find that programmer preferences do not correlate with their actual performance, motivating the need for better proxy signals. We open-source RealHumanEval to enable human-centric evaluation of new models and the study data to facilitate efforts to improve code models.

cs.SE↗

arXiv2024

Learning to Decode Collaboratively with Multiple Language Models

We propose a method to teach multiple large language models (LLM) to collaborate by interleaving their generations at the token level. We model the decision of which LLM generates the next token as a latent variable. By optimizing the marginal likelihood of a training set under our latent variable model, the base LLM automatically learns when to generate itself and when to call on one of the ``assistant'' language models to generate, all without direct supervision. Token-level collaboration during decoding allows for a fusion of each model's expertise in a manner tailored to the specific task at hand. Our collaborative decoding is especially useful in cross-domain settings where a generalist base LLM learns to invoke domain expert models. On instruction-following, domain-specific QA, and reasoning tasks, we show that the performance of the joint system exceeds that of the individual models. Through qualitative analysis of the learned latent decisions, we show models trained with our method exhibit several interesting collaboration patterns, e.g., template-filling. Our code is available at https://github.com/clinicalml/co-llm.

cs.CL↗

arXiv2024

Seq-to-Final: A Benchmark for Tuning from Sequential Distributions to a Final Time Point

Distribution shift over time occurs in many settings. Leveraging historical data is necessary to learn a model for the last time point when limited data is available in the final period, yet few methods have been developed specifically for this purpose. In this work, we construct a benchmark with different sequences of synthetic shifts to evaluate the effectiveness of 3 classes of methods that 1) learn from all data without adapting to the final period, 2) learn from historical data with no regard to the sequential nature and then adapt to the final period, and 3) leverage the sequential nature of historical data when tailoring a model to the final period. We call this benchmark Seq-to-Final to highlight the focus on using a sequence of time periods to learn a model for the final time point. Our synthetic benchmark allows users to construct sequences with different types of shift and compare different methods. We focus on image classification tasks using CIFAR-10 and CIFAR-100 as the base images for the synthetic sequences. We also evaluate the same methods on the Portraits dataset to explore the relevance to real-world shifts over time. Finally, we create a visualization to contrast the initializations and updates from different methods at the final time step. Our results suggest that, for the sequences in our benchmark, methods that disregard the sequential structure and adapt to the final time point tend to perform well. The approaches we evaluate that leverage the sequential nature do not offer any improvement. We hope that this benchmark will inspire the development of new algorithms that are better at leveraging sequential historical data or a deeper understanding of why methods that disregard the sequential nature are able to perform well.

cs.LG↗

arXiv2024

A Data-Centric Approach To Generate Faithful and High Quality Patient Summaries with Large Language Models

Patients often face difficulties in understanding their hospitalizations, while healthcare workers have limited resources to provide explanations. In this work, we investigate the potential of large language models to generate patient summaries based on doctors' notes and study the effect of training data on the faithfulness and quality of the generated summaries. To this end, we release (i) a rigorous labeling protocol for errors in medical texts and (ii) a publicly available dataset of annotated hallucinations in 100 doctor-written and 100 generated summaries. We show that fine-tuning on hallucination-free data effectively reduces hallucinations from 2.60 to 1.55 per summary for Llama 2, while preserving relevant information. We observe a similar effect on GPT-4 (0.70 to 0.40), when the few-shot examples are hallucination-free. We also conduct a qualitative evaluation using hallucination-free and improved training data. We find that common quantitative metrics do not correlate well with faithfulness and quality. Finally, we test GPT-4 for automatic hallucination detection, which clearly outperforms common baselines.

cs.CL↗

arXiv2024

Prediction-powered Generalization of Causal Inferences

Causal inferences from a randomized controlled trial (RCT) may not pertain to a target population where some effect modifiers have a different distribution. Prior work studies generalizing the results of a trial to a target population with no outcome but covariate data available. We show how the limited size of trials makes generalization a statistically infeasible task, as it requires estimating complex nuisance functions. We develop generalization algorithms that supplement the trial data with a prediction model learned from an additional observational study (OS), without making any assumptions on the OS. We theoretically and empirically show that our methods facilitate better generalization when the OS is high-quality, and remain robust when it is not, and e.g., have unmeasured confounding.

stat.ML↗

arXiv2024

Theoretical Analysis of Weak-to-Strong Generalization

Strong student models can learn from weaker teachers: when trained on the predictions of a weaker model, a strong pretrained student can learn to correct the weak model's errors and generalize to examples where the teacher is not confident, even when these examples are excluded from training. This enables learning from cheap, incomplete, and possibly incorrect label information, such as coarse logical rules or the generations of a language model. We show that existing weak supervision theory fails to account for both of these effects, which we call pseudolabel correction and coverage expansion, respectively. We give a new bound based on expansion properties of the data distribution and student hypothesis class that directly accounts for pseudolabel correction and coverage expansion. Our bounds capture the intuition that weak-to-strong generalization occurs when the strong model is unable to fit the mistakes of the weak teacher without incurring additional error. We show that these expansion properties can be checked from finite data and give empirical evidence that they hold in practice.

cs.LG↗