The FoMER benchmark is now public, evaluating multimodal models on step‑by‑step robot reasoning. Top LMMs describe scenes well but often miss safety constraints. https://getnews.me/foundation-model-embodied-reasoning-benchmark-reveals-ai-limits/ #foundationmodel #embodiedreasoning

*You never tell an LLM robot what "depth" or "3D" is, you just give the system three different mobile cameras and it emergently triangulates everything #weird #llmbot #embodiedreasoning