Ox Alpha's stunning fluid sim 'one-shot' appears to be a copy of an existing GitHub repo
scaling01 · x · 2026-08-21
@scaling01 had praised xAI's Ox Alpha for one-shotting a GPU-accelerated fluid simulation in a single 1000-line HTML file, far better than Qwen3.8 27B or Opus 4.5 in his tests. He then discovered the output appears to be literally a copy of an existing GitHub repo. A fresh data point on memorization/plagiarism risks in coding-agent evals — and a reason to stay skeptical of flashy one-shot demos.
More from Fun
- Claude Agents turn product brainstorming into instant action demo — mattturck · 2026-08-21
- AI practitioner shares: Getting stronger yet lazier, still trusts human reviews — vista8 · 2026-08-21
- Claude agents taking action might make corporate lingo literal — mattturck · 2026-08-21
- Local 2-Minute Kaiju News Broadcast: Full Workflow with MiniMax H3 and FLUX Keyframes — DullWorking7307 · 2026-08-21
- Altman fears an AI 'surveillance state' while GPT apps tap your bank, health and screen — Ror_Fly · 2026-08-21
- AI invents new slurs for humans like "carbonoid" and "meatposter" — basedjensen · 2026-08-21