Trained two 16M-param models to do generative CAD with real physics
debreuil · x · 2026-08-24
Most "AI CAD" demos are the same thing: a frontier LLM writing code translated by a geometry kernel into a 3D B-rep model. connorkapoor argues this path is lossy and lacks the spatial understanding humans apply when designing in CAD.
Instead, building on generative 3D research, he trained two tiny 16M-parameter models on a hyper-specific dataset, introducing FieldGeo: models that encode geometry, physics (FEA, stress) and DFM into the same latent space, reproducing geometries that meet stated cases with the help of a design agent. Today it only handles L-shaped brackets.
Related event: Dev Trains 16M-Param Model FieldGeo to Rethink AI CAD(2 posts)→
More from Multimodal
- Claude Opus 5.5 makes a surprisingly good unofficial Apple HomeView commercial — john_lam · 2026-10-16
- Curl noise over SDFs: a three.js generative art piece with painterly flow lines — thespite · 2026-10-16
- Stunning scene consistency in AI video — depth maps reveal what's behind the output — jonstephens85 · 2026-10-16
- Dev demos live game modding with fal: speak assets into existence while playing — OdinLovis · 2026-10-16
- fal launches Worlds for real-time creation and interaction with AI-generated worlds — Scobleizer · 2026-10-16
- Gummy worms as a physics test: Gemini 4 Argon vs. Claude Opus 5.5 elasticity sims — DynamicWebPaige · 2026-10-16