I don't think anything close to the compute needed to train chatgpt or any of the other LLM's would be possible in 1985 or 2000