NVIDIA has detailed a workflow that fine-tunes its Cosmos 3 Nano world foundation model for a video question-answering task in under a day, lifting accuracy from a 54.41% zero-shot baseline to 93.35% with minimal manual intervention. The company published the technical walkthrough on its developer blog, framing agent-driven fine-tuning as a way to solve the "last-mile accuracy" problem that has slowed real-world deployment of physical AI models.
Continue reading
The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in✓ Signed in — this article isn’t included in your current plan.