Linus Torvalds used Claude Code to build a Linux page cache fix that hit 23x read throughput, but kernel reviewers found a race before anything merged.
This happens when an LLM reads a bunch of pages with stuff it doesn’t have much internal knowledge about, and then tries to summarize that into an article.
Books contain knowledge, I doubt you will find many English scholars that disagree with that. In the same way, LLMs contain knowledge as well, they just also have the ability to output that knowledge in a more dynamic way than a book. You can argue some words that are more anthropomorphic, but knowledge has been attributed to inanimate objects for millennia.
No, LLMs only contain a bunch of numbers that predict which word is likely to follow the previous ones. Their output resembles knowledge simply because the most likely combination of words is one meant to convey information.
And books just contain a series of words. You are playing semantics games. An ebook is just a bunch of electrons shoved into a rock. Yet, somehow, it also contains knowledge.
In a less abstract way, LLMs giant relational databases that are expressed by using the relationship between previous words to select a word that might follow. It’s in the relational graph that the knowledge is stored. The information is there, and the way we’ve figured out best to get it out is by generating tokens.
LLMs do not have knowledge about anything.
Books contain knowledge, I doubt you will find many English scholars that disagree with that. In the same way, LLMs contain knowledge as well, they just also have the ability to output that knowledge in a more dynamic way than a book. You can argue some words that are more anthropomorphic, but knowledge has been attributed to inanimate objects for millennia.
No, LLMs only contain a bunch of numbers that predict which word is likely to follow the previous ones. Their output resembles knowledge simply because the most likely combination of words is one meant to convey information.
And books just contain a series of words. You are playing semantics games. An ebook is just a bunch of electrons shoved into a rock. Yet, somehow, it also contains knowledge.
In a less abstract way, LLMs giant relational databases that are expressed by using the relationship between previous words to select a word that might follow. It’s in the relational graph that the knowledge is stored. The information is there, and the way we’ve figured out best to get it out is by generating tokens.
Tell me you don’t know how deep learning works without telling me you don’t know how deep learning works.
Tell me you have zero education in epistemology without telling me you have zero education in epistemology.