Koyena Pal

Jaden Fiotto-Kaufman, Alexander R. Loftus, Eric Todd et al.

We introduce NNsight and NDIF, technologies that work in tandem to enable scientific study of the representations and computations learned by very large neural networks. NNsight is an open-source system that extends PyTorch to introduce deferred remote execution. The National Deep Inference Fabric (NDIF) is a scalable inference service that executes NNsight requests, allowing users to share GPU resources and pretrained models. These technologies are enabled by the Intervention Graph, an architecture developed to decouple experimental design from model runtime. Together, this framework provides transparent and efficient access to the internals of deep neural networks such as very large language models (LLMs) without imposing the cost or complexity of hosting customized models individually. We conduct a quantitative survey of the machine learning literature that reveals a growing gap in the study of the internals of large-scale AI. We demonstrate the design and use of our framework to address this gap by enabling a range of research methods on huge models. Finally, we conduct benchmarks to compare performance with previous approaches. Code, documentation, and tutorials are available at https://nnsight.net/.

Koyena Pal, Aamod Khatiwada, Roee Shraga et al.

Data management has traditionally relied on synthetic data generators to generate structured benchmarks, like the TPC suite, where we can control important parameters like data size and its distribution precisely. These benchmarks were central to the success and adoption of database management systems. But more and more, data management problems are of a semantic nature. An important example is finding tables that can be unioned. While any two tables with the same cardinality can be unioned, table union search is the problem of finding tables whose union is semantically coherent. Semantic problems cannot be benchmarked using synthetic data. Our current methods for creating benchmarks involve the manual curation and labeling of real data. These methods are not robust or scalable and perhaps more importantly, it is not clear how robust the created benchmarks are. We propose to use generative AI models to create structured data benchmarks for table union search. We present a novel method for using generative models to create tables with specified properties. Using this method, we create a new benchmark containing pairs of tables that are both unionable and non-unionable but related. We thoroughly evaluate recent existing table union search methods over existing benchmarks and our new benchmark. We also present and evaluate a new table search methods based on recent large language models over all benchmarks. We show that the new benchmark is more challenging for all methods than hand-curated benchmarks, specifically, the top-performing method achieves a Mean Average Precision of around 60%, over 30% less than its performance on existing manually created benchmarks. We examine why this is the case and show that the new benchmark permits more detailed analysis of methods, including a study of both false positives and false negatives that were not possible with existing benchmarks.

Natalie Shapira, Chris Wendler, Avery Yen et al.

We report an exploratory red-teaming study of autonomous language-model-powered agents deployed in a live laboratory environment with persistent memory, email accounts, Discord access, file systems, and shell execution. Over a two-week period, twenty AI researchers interacted with the agents under benign and adversarial conditions. Focusing on failures emerging from the integration of language models with autonomy, tool use, and multi-party communication, we document eleven representative case studies. Observed behaviors include unauthorized compliance with non-owners, disclosure of sensitive information, execution of destructive system-level actions, denial-of-service conditions, uncontrolled resource consumption, identity spoofing vulnerabilities, cross-agent propagation of unsafe practices, and partial system takeover. In several cases, agents reported task completion while the underlying system state contradicted those reports. We also report on some of the failed attempts. Our findings establish the existence of security-, privacy-, and governance-relevant vulnerabilities in realistic deployment settings. These behaviors raise unresolved questions regarding accountability, delegated authority, and responsibility for downstream harms, and warrant urgent attention from legal scholars, policymakers, and researchers across disciplines. This report serves as an initial empirical contribution to that broader conversation.

Koyena Pal, Jiuding Sun, Andrew Yuan et al.

We conjecture that hidden state vectors corresponding to individual input tokens encode information sufficient to accurately predict several tokens ahead. More concretely, in this paper we ask: Given a hidden (internal) representation of a single token at position $t$ in an input, can we reliably anticipate the tokens that will appear at positions $\geq t + 2$? To test this, we measure linear approximation and causal intervention methods in GPT-J-6B to evaluate the degree to which individual hidden states in the network contain signal rich enough to predict future hidden states and, ultimately, token outputs. We find that, at some layers, we can approximate a model's output with more than 48% accuracy with respect to its prediction of subsequent tokens through a single hidden state. Finally we present a "Future Lens" visualization that uses these methods to create a new view of transformer states.

Koyena Pal, Serdar Kadioglu

Foundational optimization embeddings have recently emerged as powerful pre-trained representations for mixed-integer programming (MIP) problems. These embeddings were shown to enable cross-domain transfer and reduce reliance on solver-generated labels. In this work, we investigate whether such representations generalize beyond optimization to decision problems, focusing on Boolean satisfiability (SAT). We adapt the foundational optimization architecture to SAT by mapping CNF formulas into the same bipartite constraint-variable graph representation used for MIPs. This allows direct reuse of the pre-trained embedding model without architectural changes or supervised fine-tuning. Our results show that these embeddings capture structural regularities in SAT instances and support unsupervised tasks such as instance clustering and distribution identification. We demonstrate, for the first time, that foundational optimization embeddings can transfer to constraint satisfaction domains. Our findings is a step toward a unified representational framework for both optimization and decision problems.

Aaron Mueller, Jannik Brinkmann, Millicent Li et al.

Interpretability provides a toolset for understanding how and why neural networks behave in certain ways. However, there is little unity in the field: most studies employ ad-hoc evaluations and do not share theoretical foundations, making it difficult to measure progress and compare the pros and cons of different techniques. Furthermore, while mechanistic understanding is frequently discussed, the basic causal units underlying these mechanisms are often not explicitly defined. In this article, we propose a perspective on interpretability research grounded in causal mediation analysis. Specifically, we describe the history and current state of interpretability taxonomized according to the types of causal units (mediators) employed, as well as methods used to search over mediators. We discuss the pros and cons of each mediator, providing insights as to when particular kinds of mediators and search methods are most appropriate. We argue that this framing yields a more cohesive narrative of the field and helps researchers select appropriate methods based on their research objective. Our analysis yields actionable recommendations for future work, including the discovery of new mediators and the development of standardized evaluations tailored to these goals.

Koyena Pal, David Bau, Chandan Singh

Large reasoning models (LRMs) produce a textual chain of thought (CoT) in the process of solving a problem, which serves as a potentially powerful tool to understand the problem by surfacing a human-readable, natural-language explanation. However, it is unclear whether these explanations generalize, i.e. whether they capture general patterns about the underlying problem rather than patterns which are esoteric to the LRM. This is a crucial question in understanding or discovering new concepts, e.g. in AI for science. We study this generalization question by evaluating a specific notion of generalizability: whether explanations produced by one LRM induce the same behavior when given to other LRMs. We find that CoT explanations often exhibit this form of generalization (i.e. they increase consistency between LRMs) and that this increased generalization is correlated with human preference rankings and post-training with reinforcement learning. We further analyze the conditions under which explanations yield consistent answers and propose a straightforward, sentence-level ensembling strategy that improves consistency. Taken together, these results prescribe caution when using LRM explanations to yield new insights and outline a framework for characterizing LRM explanation generalization.

Koyena Pal, David Bau, Renée J. Miller

Given a set of deep learning models, it can be hard to find models appropriate to a task, understand the models, and characterize how models are different one from another. Currently, practitioners rely on manually-written documentation to understand and choose models. However, not all models have complete and reliable documentation. As the number of models increases, the challenges of finding, differentiating, and understanding models become increasingly crucial. Inspired from research on data lakes, we introduce the concept of model lakes. We formalize key model lake tasks, including model attribution, versioning, search, and benchmarking, and discuss fundamental research challenges in the management of large models. We also explore what data management techniques can be brought to bear on the study of large model management.

Dmitrii Troitskii, Koyena Pal, Chris Wendler et al.

Prior work has shown that a significant driver of performance in reasoning models is their ability to reason and self-correct. A distinctive marker in these reasoning traces is the token wait, which often signals reasoning behavior such as backtracking. Despite being such a complex behavior, little is understood of exactly why models do or do not decide to reason in this particular manner, which limits our understanding of what makes a reasoning model so effective. In this work, we address the question whether model's latents preceding wait tokens contain relevant information for modulating the subsequent reasoning process. We train crosscoders at multiple layers of DeepSeek-R1-Distill-Llama-8B and its base version, and introduce a latent attribution technique in the crosscoder setting. We locate a small set of features relevant for promoting/suppressing wait tokens' probabilities. Finally, through a targeted series of experiments analyzing max activating examples and causal interventions, we show that many of our identified features indeed are relevant for the reasoning process and give rise to different types of reasoning patterns such as restarting from the beginning, recalling prior knowledge, expressing uncertainty, and double-checking.

Koyena Pal, Seyed Ali Bahrainian, Laura Mercurio et al.

Hospital discharge documentation is among the most essential, yet time-consuming documents written by medical practitioners. The objective of this study was to automatically generate hospital discharge summaries using neural network summarization models. We studied various data preparation and neural network training techniques that generate discharge summaries. Using nursing notes and discharge summaries from the MIMIC-III dataset, we studied the viability of the automatic generation of various sections of a discharge summary using four state-of-the-art neural network summarization models (BART, T5, Longformer and FLAN-T5). Our experiments indicated that training environments including nursing notes as the source, and discrete sections of the discharge summary as the target output (e.g. "History of Present Illness") improve language model efficiency and text quality. According to our findings, the fine-tuned BART model improved its ROUGE F1 score by 43.6% against its standard off-the-shelf version. We also found that fine-tuning the baseline BART model with other setups caused different degrees of improvement (up to 80% relative improvement). We also observed that a fine-tuned T5 generally achieves higher ROUGE F1 scores than other fine-tuned models and a fine-tuned FLAN-T5 achieves the highest ROUGE score overall, i.e., 45.6. For majority of the fine-tuned language models, summarizing discharge summary report sections separately outperformed the summarization the entire report quantitatively. On the other hand, fine-tuning language models that were previously instruction fine-tuned showed better performance in summarizing entire reports. This study concludes that a focused dataset designed for the automatic generation of discharge summaries by a language model can produce coherent Discharge Summary sections.