March prediction rings true: LLMs can now directly edit binaries
xeophon · x · 2026-09-06
Noah Ziems revisits a non-consensus prediction @xeophon made in mid-March: language models would become incredibly good at directly editing binaries. It has now come true — "binaries are basically editable code" — and Ziems argues the implications are very large, noting such high-impact, non-consensus AI predictions are rare.
- The prediction: models directly editing compiled binaries
- Status: validated in practice
- Implication: reverse-engineering-style binary patching becomes model-accessible
More from Models
- Unverified: GPT-6 Astra reportedly scores 91.8% on SpatialBench vs 80% human baseline — VraserX · 2026-09-06
- DeepSeek v4 models reportedly 30% off via third-party channel, unverified — matlabulous · 2026-09-06
- One prompt reportedly got GPT-6 Astra to build a full game and record a video — yungcontent · 2026-09-06
- One year of model progress: from broken sprite animations to fully rigged playable games — Dimillian · 2026-09-06
- AtCoder Founder's Private Bench: Astra Leads Fable5.1 by Roughly Two Generations — i_dg23 · 2026-09-06
- Ollama CEO: open models will carry 80-90% of enterprise tokens at just 10-20% of cost — victor_explore · 2026-09-06