Lucas Dixon
Lucas is a director of research at Google DeepMind, leading research efforts aimed at making breakthroughs for on-device, open-weight and sovereign AI. He works on visualizing, interpreting, and controlling these systems. Previously, he co-led PAIR (People and AI Research), advancing the capabilities to interact, control and understand LLMs. Lucas was also Chief Scientist at Jigsaw, where led work on digital security, machine learning, and data visualization, including systems focused on protecting against denial-of-service attacks, tracking global cyberattacks, documenting events in conflict zones, and addressing online toxicity. Before joining Google, he completed his PhD and worked at a university on automating mathematical reasoning and applying graphical languages to quantum information.
Research Areas
-
Algorithms and Theory
-
Data Management
-
Data Mining and Modeling
-
Distributed Systems and Parallel Computing
-
General Science
-
Human-Computer Interaction and Visualization
-
Information Retrieval and the Web
-
Machine Intelligence
-
Natural Language Processing
-
Networking
-
Security, Privacy and Abuse Prevention
-
Software Engineering
-
Software Systems
Authored Publications
Sort By
LLM Comparator: Visual Analytics for Side-by-Side Evaluation of Large Language Models
Minsuk Kahng
Michael Xieyang Liu
Krystal Kallarackal
Extended Abstracts of the CHI Conference on Human Factors in Computing Systems (CHI EA '24), ACM (2024)
Preview abstract
Automatic side-by-side evaluation has emerged as a promising approach to evaluating the quality of responses from large language models (LLMs). However, analyzing the results from this evaluation approach raises scalability and interpretability challenges. In this paper, we present LLM Comparator, a novel visual analytics tool for interactively analyzing results from automatic side-by-side evaluation. The tool supports interactive workflows for users to understand when and why a model performs better or worse than a baseline model, and how the responses from two models are qualitatively different. We iteratively designed and developed the tool by closely working with researchers and engineers at Google. This paper details the user challenges we identified, the design and development of the tool, and an observational study with participants who regularly evaluate their models.
View details
Preview abstract
An Explorable explaining the concept of patchoscopes for an external audience. Patchoscopes is an interpretability tool that allows researchers to better understand an LLMs output representations through natural language experiments.
View details
"We Need Structured Output": Towards User-centered Constraints on Large Language Model Output
Michael Xieyang Liu
Frederick Liu
Alex Fiannaca
Terry Koo
In Extended Abstract in ACM CHI Conference on Human Factors in Computing Systems (CHI EA '24), ACM (2024), pp. 9 (to appear)
Preview abstract
Large language models can produce creative and diverse responses. However, to integrate them into current developer workflows, it is essential to constrain their outputs to follow specific formats or standards. In this work, we surveyed 51 experienced industry professionals to understand the range of scenarios and motivations driving the need for output constraints from a user-centered perspective. We identified 134 concrete use cases for constraints at two levels: low-level, which ensures the output adhere to a structured format and an appropriate length, and high-level, which requires the output to follow semantic and stylistic guidelines without hallucination. Critically, applying output constraints could not only streamline the currently repetitive process of developing, testing, and integrating LLM prompts for developers, but also enhance the user experience of LLM-powered features and applications. We conclude with a discussion on user preferences and needs towards articulating intended constraints for LLMs, alongside an initial design for a constraint prototyping tool.
View details
Parameter Efficient Reinforcement Learning from Human Feedback
Hakim Sidahmed
Alex Hutcheson
Zhuonan Lin
Zhang Chen
Zac Yu
Jarvis Jin
Simral Chaudhary
Roman Komarytsia
Christiane Ahlheim
Yonghao Zhu
Bowen Li
Jessica Hoffmann
Hassan Mansoor
Wei Li
Abhinav Rastogi
2024
Preview abstract
While Reinforcement Learning from Human Feedback (RLHF) effectively aligns pretrained Large Language Models (LLMs) with human preferences, its computational cost and complexity hinder wider adoption.
This work introduces Parameter-Efficient Reinforcement Learning (PERL): by leveraging Low-Rank Adaptation (LoRA) \citep{hu2021lora} for reward model training and reinforcement learning, we are able to perform RL loops while updating only a fraction of the parameters required by traditional RLHF.
We demonstrate that the effectiveness of this method is not confined to a specific task. We compare PERL to conventional fine-tuning (full-tuning) across X highly diverse tasks, spanning from summarization to X and X, for a total of X different benchmarks - including two novel preference datasets released with this paper. Our findings show that PERL achieves comparable performance to RLHF while significantly reducing training time (up to 2x faster for reward models and 15\% faster for RL loops), and memory footprint (up to 50\% reduction for reward models and 25\% for RL loops). Finally, we provide a single set of parameters that achieves results on par with RLHF on every task, which shows the accessibility of the method.
By mitigating the computational cost and the burden of hyperparameter search, PERL facilitates broader adoption of RLHF as an LLM alignment technique.
View details
Preview abstract
An Explorable explaining the concept of patchoscopes for an external audience. Patchoscopes is an interpretability tool that allows researchers to better understand an LLMs output representations through natural language experiments.
View details
Who’s asking? User personas and the mechanics of latent misalignment
Ann Yuan
Marius Guerard
Michael Lepori
NeurIPS 2024 (2024)
Preview abstract
Why do models respond to harmful queries in some cases but not others?
Despite significant investments in improving model safety, it has been shown that misaligned capabilities remain latent in safety-tuned models. In this work, we shed light on the mechanics of this phenomenon. First, we show that even when model generations are safe, harmful content persists in hidden representations, and this content can be extracted by decoding from earlier layers. Then, we show that whether the model divulges such content depends significantly on who it is talking to, which we refer to as user persona. We study both natural language prompting and activation steering as methods for manipulating inferred user persona and show that the latter is significantly more effective at bypassing safety filters. In fact, we find it is even more effective than direct attempts to control a model's refusal tendency.
This suggests when it comes to deciding whether to respond to harmful queries, the model is deeply biased with respect to user persona. We leverage the generative capabilities of the language model itself to investigate why certain personas break model safeguards, and discover that they enable the model to form more charitable interpretations of otherwise dangerous queries. Finally, we show that we can predict a persona’s effect on refusal given only the geometry of its steering vector.
View details
Preview abstract
A common method to study deep learning systems is to create simplified representations---for example, using singular value decomposition to visualize the model's hidden states in a lower dimensional space. This approach assumes that the simplified model is faithful to the original model. Here, we illustrate an important caveat to this assumption: even if a simplified representation of the model can accurately approximate the original model on the training set, it may fail to match its behavior out of distribution; the understanding developed from simplified representations may be an illusion. We illustrate this by training Transformer models on controlled datasets with systematic generalization splits, focusing on the Dyck balanced-parenthesis languages. We simplify these models using tools like dimensionality-reduction and clustering, and find clear patterns in the resulting representations. We then explicitly test how these simplified proxy models match the original models behavior on various out-of-distribution test sets. Generally, the simplified proxies are less faithful out of distribution. For example, in cases where the original model generalizes to novel structures or deeper depths, the simplified model may fail to generalize, or may generalize too well. We then show the generality of these results: even model simplifications that do not directly use data can be less faithful out of distribution, and other tasks can also yield generalization gaps. Our experiments raise questions about the extent to which mechanistic interpretations derived using tools like SVD can reliably predict what a model will do in novel situations.
View details
Large Language Models are Competitive Near Cold-start Recommenders for Language- and Item-based Preferences
Scott Sanner
Proceedings of ACM Conference on Recommender Systems (RecSys ’23) (2023)
Preview abstract
Traditional recommender systems leverage users' item preference history to recommend novel content that users may like. However, dialog interfaces that allow users to express language-based preferences offer a fundamentally different modality for preference input. Inspired by recent successes of prompting paradigms for large language models (LLMs), we study their use for making recommendations from both item-based and language-based preferences in comparison to state-of-the-art item-based collaborative filtering (CF) methods. To support this investigation, we collect a new dataset consisting of both item-based and language-based preferences elicited from users along with their ratings on a variety of (biased) recommended items and (unbiased) random items. Among numerous experimental results, we find that LLMs provide competitive recommendation performance for pure language-based preferences (no item preferences) in the near cold-start case in comparison to item-based CF methods, despite having no supervised training for this specific task (zero-shot) or only a few labels (few-shot). This is particularly promising as language-based preference representations are more explainable and scrutable than item-based or vector-based representations.
View details
Do machine learning models memorize or generalize?
Adam Pearce
IEEE VISxAI (2023)
Preview abstract
In 2021, researchers made a striking discovery while training a series of tiny models on toy tasks [1]. They found a set of models that suddenly flipped from memorizing their training data to correctly generalizing on unseen inputs after training for much longer. This phenomenon – where generalization seems to happen abruptly and long after fitting the training data – is called grokking and has sparked a flurry of interest [2, 3, 4, 5, 6].
Do more complex models also suddenly generalize after they’re trained longer? Large language models can certainly seem like they have a rich understanding of the world, but they might just be regurgitating memorized bits of the enormous amount of text they’ve been trained on [7, 8]. How can we tell if they’re generalizing or memorizing?
In this article we’ll examine the training dynamics of a tiny model and reverse engineer the solution it finds – and in the process provide an illustration of the exciting emerging field of mechanistic interpretability [9, 10]. While it isn’t yet clear how to apply these techniques to today’s largest models, starting small makes it easier to develop intuitions as we progress towards answering these critical questions about large language models.
View details
