How long before I can use an open-source LLM roughly as easily as I can use chatGPT that is of comparable quality as 3.5?