Context Vim idea: diffusion LLMs could edit full code pages with control tokens
ryunuck · x · 2026-11-14
Replying to the "introduce Claude to vim" joke, @ryunuck argues Context Vim should start with diffusion LLMs: since the context window doesn't need sequential generation, you can inject a full page of code and edit across it, adding special control tokens like scroll tokens trained with RL. He quips that reaching this point would make models "truly AGI — but it's still not AGI."
More from coding & agent
- Google details /teamwork: long-horizon agents behind its research breakthroughs — ymatias · 2026-11-14
- Google's Antigravity multiagent proves 2013 info-theory conjecture in Lean 4 — ymatias · 2026-11-14
- Dev finds asking LLMs for prompt and tool design advice often hurts results — mimi10v3 · 2026-11-14
- Yohei Nakajima to headline 'Agents on the Frontier' Pro Series live deep dive next Tuesday — ColleenMBrady · 2026-11-14
- Yogurt: open-source local-first Granola alternative for private AI meeting notes — supportingthedogs · 2026-11-14
- Anthropic Breaks Down the Five Primitives Behind Always-On AI Agents — dotey · 2026-11-14