phantom-kv removes LLM refusals via hot-swappable KV cache, no weight edits needed
cyb3rops · x · 2026-09-22
Developer lordx64 shipped phantom-kv, a refusal-removal system for LLMs that touches zero weights. Instead of editing the model, it loads a small learned bank of key/value tensors into the model's KV cache as context.
The effect turns 'uncensoring' from a permanent checkpoint edit into a per-request, hot-swappable capability mode: unload the cache and the base model is byte-identical again. A /pill {blue, red, black} demo is included in the thread.
Related event: phantom-kv removes LLM refusals via 18MB KV cache injection(3 posts)→
More from Models
- Gemini 4 Carbon reportedly drops as builders share 10 early wild examples — socialwithaayan · 2026-11-12
- How to cut Fable 5.2 token costs by 90%: Arc routing plus a lean CLAUDE.md — EXM7777 · 2026-11-12
- Partial RuneBench results: Inkling 2 Small scores 5.314, far above original — mariofilhoml · 2026-11-12
- Grok Bot 1.0 Announced: Shareable Bots With Upgraded Collaboration — SpaceXSI · 2026-11-12
- Polymarket odds put Chinese AI lab at #1 by year-end at just 6% — Polymarket · 2026-11-12
- FutureHouse's deslopper trains Kosmos to write indistinguishably from human experts — proteinrosh · 2026-11-12