#interpretability

Token-level entropy trajectories from LLM logprobs. Models can measure their own uncertainty — grounding it in truth requires external verification.

Supporting files for Anoikis², Case Studies in GPT-2 Small/EVE Frontier (Gamified Mechanistic Interpretability)

Prior- and constraint-driven directed-evolution agent on the GB1 four-site landscape: seven-arm controlled comparison, auditable constraints, and an honest null result for the LLM

Does an LLM's willingness to blackmail depend on its persona — and on whether that persona is prompted or trained into the weights? Two-part study on Gemma-3-12B: 16 prompted personas (n=100) + 10 QLoRA persona adapters, in the Lynch et al. shutdown scenario.