PaLM: Scaling Language Modeling with Pathways
Scaling language models using Google's Pathways system, achieving state-of-the-art performance across hundreds of language understanding and generation tasks.
My research focuses on advancing the capabilities of large language models and on the computational principles of intelligence. I have publications and patents spanning state of the art AI, machine learning, human-computer interaction, and quantitative finance.
Citation Growth
35,000+ citations
Sort by:
Evaluating Large Language Models Trained on Code
Introduced HumanEval benchmark for evaluating code generation capabilities of large language models, foundational work for GitHub Copilot and similar tools.
Gemini: A Family of Highly Capable Multimodal Models
Technical report on Gemini, Google DeepMind's multimodal AI model family with state-of-the-art capabilities across text, image, audio, and video understanding.
Gemini 1.5: Unlocking Multimodal Understanding Across Millions of Tokens
Long-context understanding with up to 10M token context window, enabling new applications in document analysis and reasoning.
PaLM 2 Technical Report
Next generation of PaLM with improved multilingual, reasoning, and coding capabilities.
Beyond the Imitation Game: Quantifying and Extrapolating the Capabilities of Language Models
Comprehensive benchmark (BIG-bench) with over 200 tasks for evaluating language model capabilities beyond simple imitation.
Solving Quantitative Reasoning Problems with Language Models
Demonstrated how language models can solve complex mathematical and quantitative reasoning problems through improved training approaches.
Gemini 2.5: Pushing the Frontier with Advanced Reasoning and Agentic Capabilities
Latest generation Gemini model with enhanced reasoning, multimodality, and agentic capabilities.
Gemma 3 Technical Report
Open-source language model family designed for responsible AI development and deployment.
Grokking: Generalization Beyond Overfitting on Small Algorithmic Datasets
Discovered the "grokking" phenomenon where neural networks suddenly generalize long after overfitting, with implications for understanding deep learning.
Exploring Length Generalization in Large Language Models
Analysis of how language models generalize to longer sequences than seen during training.
Rational Orbits Around Charged Black Holes
Classification of eccentric timelike orbits in charged black hole spacetime using dynamical systems theory, with applications to gravitational wave astronomy.
Methods and Systems for Automated Generation of Personalized Messages
Patent for systems and methods to automatically generate personalized content for sales and marketing communications using machine learning.
Method and Apparatus for Dynamic Information Visualization
Patent for methods and systems for dynamic visualization of complex information and data patterns.
Evidence of Market Manipulation in the Financial Crisis
Statistical analysis revealing evidence of market manipulation ("bear raids") at the beginning of the 2007 financial crisis.
Artificial Intelligence and Complex Statistical Modeling in Glaucoma Diagnosis
Application of AI and statistical methods to improve diagnosis and treatment of glaucoma.
Black Box Attacks on Transformer Language Models
Research on adversarial attacks against transformer-based language models.
Vulnerability Analysis of High Dimensional Complex Systems
Analysis of vulnerability patterns in complex high-dimensional systems using network theory and statistical methods.
Regulation of Short Selling: The Uptick Rule and Market Stability
Analysis of short selling regulations and their impact on market stability, presented to the SEC.