colored-dye's blog

Interpreting the minds of machines and myself


Scaling Properties of Same-Family On-Policy Distillation

how much RL-acquired capability transfers across model scales via on-policy distillation, how fast, and how to predict it before training.

17 min read   ·   2026

Training Prompt-only Steering Vectors in a Principled Manner

our recent work on prompt-only SV and SV training dynamics.

13 min read   ·   2026

Concept Distributed Alignment Search for Faithful Representation Steering

discussions regarding our recent work on faithful representation steering.

20 min read   ·   2026