There's nothing new here, but the U.S. AI industry had a conniption. Dividing tasks into functions is simple parallel processing. That's what DeepSeek did. It parsed the LLM task into pieces, and built expertise around each separate piece. Then it gave what it did away. I think the giving away, the open source bit, is what really freaked people out.
It's amazing in retrospect how quickly the innovation around OpenAI became a land grab, and how quickly open source shut all that down, through DeepSeek. Someone give Eric Raymond a cigar.
There's nothing new here, but the U.S. AI industry had a conniption. Dividing tasks into functions is simple parallel processing. That's what DeepSeek did. It parsed the LLM task into pieces, and built expertise around each separate piece. Then it gave what it did away. I think the giving away, the open source bit, is what really freaked people out.
It's amazing in retrospect how quickly the innovation around OpenAI became a land grab, and how quickly open source shut all that down, through DeepSeek. Someone give Eric Raymond a cigar.