LUKE
SAMUEL
MARKS
DEEP LEARNING AND DEEP LEARNING ADJACENT THINGS
Email: marksluke076[at]gmail[dot]com
Publications
- Beyond Linear Steering: Unified Multi-Attribute Control for Language Models
- TinySQL: A Progressive Text-to-SQL Dataset for Mechanistic Interpretability Research
- Enhancing Neural Network Interpretability with Feature-Aligned Sparse Autoencoders
- Interpreting Learned Feedback Patterns in Large Language Models (NeurIPS 2024)
Awards
- Received a grant from Emergent Ventures in 2024.
- Received a grant from BERI in 2024.
- Became a Magnificent Grantee in 2024.
- Received a grant from Lightspeed Grants in 2023.
Fellowships
- MATS 2025 summer scholar.
- Interned at Martian in 2025.
- Was a summer intern at the Torr Vision Group in 2024.
- Was a research fellow at Apart Research in 2023.
- Became a Non-Trivial Fellow in 2023.
Other
- Dropped out of high school in 2024.
Interests
- Deep learning
- Mind uploading
- Homomorphic encryption
Projects
- BadTransformer: Simple transformer with only NumPy and Python as dependencies.
- Reward Hacking: A replication of Ibarz et al. In this project I implement common architectures in Deep RL from the time and try to replicate the reward hacking behaviors observed in the Atari game Hero.