colored_dye's picture

4

colored_dye

colored-dye

·

colored-dye

AI & ML interests

mechanistic interpretability, training data attribution, representation steering

Recent Activity

upvoted a paper 5 days ago

PragLocker: Protecting Agent Intellectual Property in Untrusted Deployments via Non-Portable Prompts

upvoted a paper 5 days ago

Faithful Bi-Directional Model Steering via Distribution Matching and Distributed Interchange Interventions

updated a dataset 5 days ago

colored-dye/concept500-contrastive

View all activity

Organizations

None yet

upvoted 2 papers 5 days ago

PragLocker: Protecting Agent Intellectual Property in Untrusted Deployments via Non-Portable Prompts

Paper • 2605.05974 • Published 9 days ago • 1

Faithful Bi-Directional Model Steering via Distribution Matching and Distributed Interchange Interventions

Paper • 2602.05234 • Published Feb 5 • 1

upvoted a paper 6 days ago

Towards Steering without Sacrifice: Principled Training of Steering Vectors for Prompt-only Interventions

Paper • 2605.05983 • Published 9 days ago • 1

upvoted a paper 11 months ago

Large Language Models are Locally Linear Mappings

Paper • 2505.24293 • Published May 30, 2025 • 14