Running NVIDIA Cosmos Locally on Mac: Gets Motion Right, Fails at Counting
MaziyarPanahi · x · 2026-08-08
A developer ran NVIDIA's Cosmos-Reason1-7B model locally on a Mac Studio (M3 Ultra) using MLX 8-bit quantization, combined with Meta SAM 3.1 for segmentation.
He designed an interesting eval: asking the model to predict the next 3 seconds of video, then rolling the actual footage to let the model grade itself. The results show the model accurately predicted the motion trajectory of jets, but hilariously failed at counting (counting 6 objects when there were only 4).
Related event: Developers Test NVIDIA Cosmos Model Locally on Mac(3 posts)→
More from Fun
- Google Intern Secures 1:1 with Jeff Dean via Cold Calendar Invite, Sparking Workplace Culture Discussion — breadli428 · 2026-08-08
- OpenAI's Quick Personality Fix Sparks Debate Over Intentional Dumbing Down — Angaisb_ · 2026-08-08
- Using AI Agents to Auto-Generate Paper Rebuttals Against Reviewers — CSProfKGD · 2026-08-08
- Connecting ChatGPT to Bank Accounts: AI Apologizes After Spending Spree — ThePeterMick · 2026-08-08
- Tech Title Meme: Everyone's a CEO on LinkedIn, Nobody Cares on X — _Stocko_ · 2026-08-08
- DeepSeek Stuck in an Infinite 'Thinking' Loop — _wOvAN_ · 2026-08-08