Fred Hohman

Understanding Annotator Decision-making with Interpretability

2026-04-28T00:00:00+00:00

Safety policies define what constitutes safe and unsafe AI outputs, guiding data annotation and model development. However, annotation disagreement is pervasive and can stem from multiple sources such as operational failures (annotators misunderstand or misexecute the task), policy ambiguity (policy wording leaves room for interpretation), or value pluralism (different annotators hold different perspectives on safety). Distinguishing these sources matters. For example, operational failures call for quality control, ambiguity calls for policy clarification, and pluralism calls for deliberation about incorporating diverse perspectives. Yet understanding why annotators disagree is difficult. Directly asking annotators for their reasoning is costly, substantially increasing annotation burden, and can be unreliable for both human and LLM annotators as self-reported reasoning often fails to reflect actual decision processes.

We introduce Annotator Policy Models (APMs), interpretable models that learn annotators’ internal safety policies from labeling behavior alone, making annotator reasoning visible and comparable without additional annotation effort. We validate that APMs accurately model annotator safety policy (>80% accuracy), faithfully predict responses to counterfactual edits, and recover known policy differences in controlled settings. Applying APMs to LLM and human annotations, we demonstrate two core applications: (1) surfacing policy ambiguity by revealing how annotators interpret safety instructions differently, and (2) surfacing value pluralism by uncovering systematic differences in safety priorities across demographic groups. Together, these capabilities support more targeted, transparent, and inclusive safety policy design.

Semantic Regexes: Auto-Interpreting LLM Features with a Structured Language

2026-04-24T00:00:00+00:00

Automated interpretability aims to translate large language model (LLM) features into human understandable descriptions. However, these natural language feature descriptions are often vague, inconsistent, and require manual relabeling. In response, we introduce semantic regexes, structured language descriptions of LLM features. By combining primitives that capture linguistic and semantic feature patterns with modifiers for contextualization, composition, and quantification, semantic regexes produce precise and expressive feature descriptions. Across quantitative benchmarks and qualitative analyses, we find that semantic regexes match the accuracy of natural language while yielding more concise and consistent feature descriptions. Moreover, their inherent structure affords new types of analyses, including quantifying feature complexity across layers, scaling automated interpretability from insights into individual features to model-wide patterns. Finally, in user studies, we find that semantic regex descriptions help people build accurate mental models of LLM feature activations.

Embedding Atlas: Low-Friction, Interactive Embedding Visualization

2025-09-26T00:00:00+00:00

Embedding projections are popular for visualizing large datasets and models. However, people often encounter “friction” when using embedding visualization tools: (1) barriers to adoption, e.g., tedious data wrangling and loading, scalability limits, no integration of results into existing workflows, and (2) limitations in possible analyses, without integration with external tools to additionally show coordinated views of metadata. In this paper, we present Embedding Atlas, a scalable, interactive visualization tool designed to make interacting with large embeddings as easy as possible. Embedding Atlas uses modern web technologies and advanced algorithms – including density-based clustering, and automated labeling – to provide a fast and rich data analysis experience at scale. We evaluate Embedding Atlas with a competitive analysis against other popular embedding tools, showing that Embedding Atlas’s feature set specifically helps reduce friction, and report a benchmark on its real-time rendering performance with millions of points. Embedding Atlas is available as open source to support future work in embedding-based analysis.

Policy Maps: Tools for Guiding the Unbounded Space of LLM Behaviors

2025-09-25T00:00:00+00:00

AI policy sets boundaries on acceptable behavior for AI models, but this is challenging in the context of large language models (LLMs): how do you ensure coverage over a vast behavior space? We introduce policy maps, an approach to AI policy design inspired by the practice of physical mapmaking. Instead of aiming for full coverage, policy maps aid effective navigation through intentional design choices about which aspects to capture and which to abstract away. With Policy Projector, an interactive tool for designing LLM policy maps, an AI practitioner can survey the landscape of model input-output pairs, define custom regions (e.g., “violence”), and navigate these regions with if-then policy rules that can act on LLM outputs (e.g., if output contains “violence” and “graphic details,” then rewrite without “graphic details”). Policy Projector supports interactive policy authoring using LLM classification and steering and a map visualization reflecting the AI practitioner’s work. In an evaluation with 12 AI safety experts, our system helps policy designers craft policies around problematic model behaviors such as incorrect gender assumptions and handling of immediate physical safety threats.

Apple Intelligence Foundation Language Models: Tech Report 2025

2025-07-18T00:00:00+00:00

We introduce two multilingual, multimodal foundation language models that power Apple Intelligence features across Apple devices and services: (i) a 3B-parameter on-device model optimized for Apple silicon through architectural innovations such as KV-cache sharing and 2-bit quantization-aware training; and (ii) a scalable server model built on a novel Parallel-Track Mixture-of-Experts PT-MoE transformer that combines track parallelism, mixture-of-experts sparse computation, and interleaved global-local attention to deliver high quality with competitive cost on Apple’s Private Cloud Compute platform. Both models are trained on large-scale multilingual and multimodal datasets sourced via responsible web crawling, licensed corpora, and high-quality synthetic data, then further refined with supervised fine-tuning and reinforcement learning on a new asynchronous platform. The resulting models support several additional languages while understanding images and executing tool calls. In public benchmarks and human evaluations, both the server model and the on-device model match or surpass comparably sized open baselines. A new Swift-centric Foundation Models framework exposes guided generation, constrained tool calling, and LoRA adapter fine-tuning, allowing developers to integrate these capabilities with a few lines of code. The latest advancements in Apple Intelligence models are grounded in our Responsible AI approach with safeguards like content filtering and locale-specific evaluation, as well as our commitment to protecting our users’ privacy with innovations like Private Cloud Compute.

A Scalable Approach to Clustering Embedding Projections

2025-04-14T00:00:00+00:00

Interactive visualization of embedding projections is a useful technique for understanding data and evaluating machine learning models. Labeling data within these visualizations is critical for interpretation, as labels provide an overview of the projection and guide user navigation. However, most methods for producing labels require clustering the points, which can be computationally expensive as the number of points grows. In this paper, we describe an efficient clustering approach using kernel density estimation in the projected 2D space instead of points. This algorithm can produce high-quality cluster regions from a 2D density map in a few hundred milliseconds, orders of magnitude faster than current approaches. We contribute the design of the algorithm, benchmarks, and applications that demonstrate the utility of the algorithm, including labeling and summarization.

Exploring Empty Spaces: Human-in-the-Loop Data Augmentation

2025-03-25T00:00:00+00:00

Data augmentation is crucial to make machine learning models more robust and safe. However, augmenting data can be challenging as it requires generating diverse data points to rigorously evaluate model behavior on edge cases and mitigate potential harms. Creating high-quality augmentations that cover these “unknown unknowns” is a time- and creativity-intensive task. In this work, we introduce Amplio, an interactive tool to help practitioners navigate “unknown unknowns” in unstructured text datasets and improve data diversity by systematically identifying empty data spaces to explore. Amplio includes three human-in-the-loop data augmentation techniques: Augment with Concepts, Augment by Interpolation, and Augment with Large Language Model. In a user study with 18 professional red teamers, we demonstrate the utility of our augmentation methods in helping generate high-quality, diverse, and relevant model safety prompts. We find that Amplio enabled red teamers to augment data quickly and creatively, highlighting the transformative potential of interactive augmentation workflows.

Compress and Compare: Interactively Evaluating Efficiency and Behavior Across ML Model Compression Experiments

2024-09-27T00:00:00+00:00

To deploy machine learning models on-device, practitioners use compression algorithms to shrink and speed up models while maintaining their high-quality output. A critical aspect of compression in practice is model comparison, including tracking many compression experiments, identifying subtle changes in model behavior, and negotiating complex accuracy-efficiency trade-offs. However, existing compression tools poorly support comparison, leading to tedious and, sometimes, incomplete analyses spread across disjoint tools. To support real-world comparative workflows, we develop an interactive visual system called Compress and Compare. Within a single interface, Compress and Compare surfaces promising compression strategies by visualizing provenance relationships between compressed models and reveals compression-induced behavior changes by comparing models’ predictions, weights, and activations. We demonstrate how Compress and Compare supports common compression analysis tasks through two case studies, debugging failed compression on generative language models and identifying compression artifacts in image classification models. We further evaluate Compress and Compare in a user study with eight compression experts, illustrating its potential to provide structure to compression workflows, help practitioners build intuition about compression, and encourage thorough analysis of compression’s effect on model behavior. Through these evaluations, we identify compression-specific challenges that future visual analytics tools should consider and Compress and Compare visualizations that may generalize to broader model comparison tasks.

Apple Intelligence Foundation Language Models

2024-07-31T00:00:00+00:00

We present foundation language models developed to power Apple Intelligence features, including a ~3 billion parameter model designed to run efficiently on devices and a large server-based language model designed for Private Cloud Compute. These models are designed to perform a wide range of tasks efficiently, accurately, and responsibly. This report describes the model architecture, the data used to train the model, the training process, how the models are optimized for inference, and the evaluation results. We highlight our focus on Responsible AI and how the principles are applied throughout the model development.

Biscuit: Scaffolding LLM-Generated Code with Ephemeral UIs in Computational Notebooks

2024-04-12T00:00:00+00:00

Novices frequently engage with machine learning tutorials in computational notebooks and have been adopting code generation technologies based on large language models (LLMs). However, they encounter difficulties in understanding and working with code produced by LLMs. To mitigate these challenges, we introduce a novel workflow into computational notebooks that augments LLM-based code generation with an additional ephemeral UI step, offering users UI-based scaffolds as an intermediate stage between user prompts and code generation. We present this workflow in BISCUIT, an extension for JupyterLab that provides users with ephemeral UIs generated by LLMs based on the context of their code and intentions, scaffolding users to understand, guide, and explore with LLM-generated code. Through a user study where 10 novices used BISCUIT for machine learning tutorials, we discover that BISCUIT offers user semantic representation of code to aid their understanding, reduces the complexity of prompt engineering, and creates a playground for users to explore different variables and iterate on their ideas. We discuss the implications of our findings for UI-centric interactive paradigm in code generation LLMs.