Steering a single neuron in Olmo 3 bypasses post-trained assistant behavior

Sauers_ · x · 2026-10-22

• Sharing an experiment on Olmo 3 (without continued post-training): asked to introduce itself, the model produced a long base-model-style mystic creation text ("Let there be ten thousand stars...").
• Key finding: steering a single neuron can completely bypass the post-training "assistant basin", recovering creative outputs thought to be lobotomized by post-training.
• A quoted tweet adds that unsupervised latent identity search surfaces entities like "catalpa tree" and "Chinese spirit", hinting at rich latent identity structures inside the model.

Original post →

More from Research

• Neural networks 'clean and transparent'? Turing Post weighs both sides of the interpretability debate — TheTuringPost · 2026-10-22
• Holon-S1, never trained on osteosarcoma, reproduces clinicians' molecular finding from slides — iScienceLuvr · 2026-10-22
• ATEST: load-aware test-time scaling cuts LLM inference energy up to 48x — Mobina Kashaniyan · 2026-10-22
• New tetrahedron theorem found with Claude after scanning 1.6M sphere combos — IgorCarron · 2026-10-22
• ROME attacks: poisoned Python can surgically rewrite LLM weights in GPU memory — SirJerryLion · 2026-10-22
• UC Berkeley opens joint Statistics×EECS tenure-track hire in Trustworthy AI — jasondeanlee · 2026-10-22

Research channel →