I’ve been seeing people claim that DeepSeek’s base model was trained off OpenAI, and that you can achieve DeepSeek R1 through some simpler and less expensive process. I don’t really understand it, but if LLMs are becoming good at coding and if a human can derive one LLM from another, then it would make sense that we are technically close to AI having the potential to replicate itself. But that doesn’t feel like the same as an AI have the sentience to replicate itself.