Former OpenAI researcher Andrey Karpaty has launched an open-source project called Nanochat. The project allows users to train and run their own AI models on GPU cloud servers. Training a small model takes a minimum of 4 hours and costs $100. 12 hours of training will create a GPT-2-level model, and 42 hours and $1,000 will create a more advanced AI assistant.
Karpaty has worked at OpenAI since 2015, and later led projects at Tesla and Eureka Labs. The project creates accessible opportunities for studying artificial intelligence and creating personal AI assistants.
