Ryan Heuser on Mythryl: “Different models have different displacement strategies and…”

You can literally watch repression & displacement consolidate over fine-tuning: here's next-token probs across checkpoints of OLMo-3-7B-Think-SFT. Explicit words are repressed almost instantly but safer (displaced) alternatives emerge much later. It learns what not to say before what to say instead.

Ryan Heuser@ryanheuser.com

Submitting this abstract to "Accelerationism Revisited", a symposium in Dublin. Mapping psychoanalytic topology in LLM base models → instruction-tuned → safety-tuned models. They progressively "displace" (in Freudian sense) censored content into adjacent semantics, even across hidden model layers.

Post