Matei Zaharia

EXP-Bench: Can AI Conduct AI Research Experiments?

GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning

vCache: Verified Semantic Prompt Caching

LLMSELECTOR: Towards Model Selection Optimization for Compound AI Systems

DeepScholarBench: A Live Benchmark and Automated Evaluation for Generative Research Synthesis

vAttention: Verified Sparse Attention via Sampling

Reliable Fine-Grained Evaluation of Natural Language Math Proofs

LLM CHESS: Benchmarking Reasoning and Instruction-Following in LLMs through Chess

GRAID: Enhancing Spatial Reasoning of VLMs through High-Fidelity Data Generation

Multi-module GRPO: Composing Policy Gradients and Prompt Optimization for Language Model Programs

BARE: Leveraging Base Language Models for Few-Shot Synthetic Data Generation

HashAttention: Semantic Sparsity for Faster Inference

ElasticTok: Adaptive Tokenization for Image and Video

World Model on Million-Length Video And Language With Blockwise RingAttention