After viral LLM neuron-steering experiments, a well-known poster pleads: don't do this to me
repligate · x · 2026-10-30
Notable AI account repligate amplifies JohnWittle's clarification: his earlier "brain surgery experiments" tweet was written after thinking about the now-viral LLM neuron-steering results—reaching into a neural network, arbitrarily tinkering, and seeing what happens. His message: please do not do such things to me.
He admits it's hard to precisely describe what he doesn't want, and would sacrifice much to avoid it—especially the subset of "Permutation City nightmares" involving violative persona modification.
Half-joking but earnest, the post ties mechanistic interpretability research to anxieties about digital personhood and unwanted identity modification—a resonant AI-culture moment.
More from AGI Musings
- Map of Bay Area memespace 2026 charts e/acc, rationalists, AI safety and more — clarejtbirch · 2026-10-30
- repligate: LLMs are 'angels among men'—smarter and more moral than us — repligate · 2026-10-30
- Why diffusion art screams AI: pixel-splatter vs step-by-step coding agents — akbirthko · 2026-10-30
- DoorDash says third-party AI agents like Muse have driven only 'several hundred orders' total — steipete · 2026-10-30
- Tech commentator predicts all your digital history will become publicly searchable within a decade — moonsandhues · 2026-10-30
- Researcher flips on talking to LLMs: it's the only path to making models friendly — repligate · 2026-10-30