Daniil Gavrilov

Guided Star-Shaped Masked Diffusion

Kronecker Factorization Improves Efficiency and Interpretability of Sparse Autoencoders

You Do Not Fully Utilize Transformer's Representation Capacity

The Differences Between Direct Alignment Algorithms are a Blur

ESSA: Evolutionary Strategies for Scalable Alignment

PRPO: Collaborative Online Policy Learning in Personalized RLHF

Rank-Then-Act: Reward-Free Control from Frame-Order Progress

Small Vectors, Big Effects: A Mechanistic Study of RL-Induced Reasoning via Steering Vectors

Teach Old SAEs New Domain Tricks with Boosting

Mechanistic Permutability: Match Features Across Layers

Learn Your Reference Model for Real Good Alignment

Analyze Feature Flow to Enhance Interpretation and Steering in Language Models