Nikita Balagansky

Nikita Balagansky on Hugging Face Daily Papers: 11 papers, 4 in the top 3 of their day, 426 upvotes.

1. Unstable Features, Reproducible Subspaces: Understanding Seed Dependence in Sparse Autoencoders 8 upvotes, #28 of 2026-06-16
2. Interpreting and Steering a Text-to-Speech Language Model with Sparse Autoencoders 12 upvotes, #19 of 2026-06-10
3. Trust-Region Behavior Blending for On-Policy Distillation 65 upvotes, #3 of 2026-06-01
4. Next Embedding Prediction Makes World Models Stronger 17 upvotes, #11 of 2026-03-04
5. Teach Old SAEs New Domain Tricks with Boosting 11 upvotes, #12 of 2025-07-18
6. Train Sparse Autoencoders Efficiently by Utilizing Features Correlation 21 upvotes, #16 of 2025-05-30
7. You Do Not Fully Utilize Transformer's Representation Capacity 33 upvotes, #8 of 2025-02-19
8. Analyze Feature Flow to Enhance Interpretation and Steering in Language Models 53 upvotes, #1 of 2025-02-07
9. Mechanistic Permutability: Match Features Across Layers 16 upvotes, #7 of 2024-10-14
10. Learn Your Reference Model for Real Good Alignment 75 upvotes, #1 of 2024-04-16
11. Linear Transformers with Learnable Kernel Functions are Better In-Context Models 81 upvotes, #1 of 2024-02-19

Data: hysts-bot-data/daily-papers-stats and the Daily Papers API. Open data: tardellirs/paper-pulse-data. Sister project: Model Pulse, the download history of every model on the Hub.