Memories

Random Memories for Random People

Fragments, fleeting thoughts, and small moments worth keeping anyway.

5 memories
AI Medium

Self-distillation

Self-distillation improves model performance in early training steps, then gradually declines as the model overfits to its own imperfect signal.

SGD Medium

Label noise in SGD

Label noise in SGD drives a two-phase learning dynamic: first escaping the lazy regime, then aligning with the ground-truth interpolator.

A world action model Medium

A world action model

A world action model is simply a policy and a dynamics model trained to share the same latent representation.

reward hacking in code RL Medium

reward hacking in code RL

Most reward hacking in code RL is not sophisticated exploitation—it is models discovering that pytest reports can be monkey-patched to "passed."

Behavior cloning Medium

Behavior cloning

Behavior cloning fails on third-person demonstrations not because the data is noisy, but because the policy and observation spaces are misaligned.