The CEO of USA's AI chip giant NVIDIA (NVDA.US), Jensen Huang, unveiled the Cosmos base model at the Consumer Electronics Show CES 2005, which can generate photo-realistic videos for training Siasun Robot&Automation and self-driving Automobiles, at a cost much lower than using traditional data.

By creating so-called 'synthetic' training data, the model can help Siasun Robot&Automation and Automobiles understand the physical world, just as large language models assist chatbots in generating responses in natural language. Users will be able to provide Cosmos with a text description to generate a video that conforms to the laws of physics, which is much cheaper than current data collection methods, such as placing Automobiles on the road to collect videos, or having people teach Siasun Robot&Automation repetitive tasks.

Cosmos will be provided under an 'open license', similar to Meta's (META.US) Llama 3 language model.