Testing visual intelligence: model estimates container volume within ~50ml
deliprao · x · 2026-11-21
To sanity-check a multimodal model's visual estimation, the author filled a container to a target level and measured the actual volume with a measuring jar. The model's estimate from the image came within roughly 50ml (slightly over)—a surprisingly practical use of visual intelligence in these models.
More from Fun
- 'Give Alpoge all the Fields Medals for the next 10 years,' AI math fans joke — cgarciae88 · 2026-11-21
- The irony: Cursor wants to drop Chinese open models, yet its own model is fine-tuned from Kimi — bindureddy · 2026-11-21
- littmath: AI may prove the Riemann hypothesis in 2026, yet Apple's LLM Siri lands 2027 — littmath · 2026-11-21
- Intel Fellow publicly calls out Google over employee's threatening AI-anger tweet — JFPuget · 2026-11-21
- Hobbyist's virtual worms with neural networks reproduce classic optimal foraging research — neuroecology · 2026-11-21
- Cyberpunk 3D pixel-art Gaussian splats wow a16z's Martin Casado — Kyrannio · 2026-11-21