For those unaware, Linus Torvalds this year has deeply cast his lot with the hyperscale LLMs for development of Linux code, dismissing arguments that this is an abdication of responsibility over that code base.
For those unaware, Linus Torvalds this year has deeply cast his lot with the hyperscale LLMs for development of Linux code, dismissing arguments that this is an abdication of responsibility over that code base.
Here’s the thing. What is it that an LLM does? I would argue, it sumarizes. It’s a probabilistic model of human language. Give it a corpus of text, and it can “understand” that data and extract “meaning” from it. Is that useful? Totally. For a number of uses. Many coding tasks, research, translation, etc are really an effort in summarization. Is it going to solve every problem humans ever had? I doubt it. Language is the human model of reality, and we have a tendency to equivocate our model with the real thing. We’re just THAT hard wired for words. I think the hype and the marketing has overshadowed what I think is a very useful application for machine learning. I think Torvalds is overall correct. I feel his position is inconvenient for political reasons. I think that open source stands in stark contrast to the world the big tech “AI” class want to build. But I don’t really think he’s wrong perse.