> Pretty much everyone agrees that these models, no matter how good they are at writing, don’t understand what they are writing, and can’t develop new ideas, on

> Pretty much everyone agrees that these models, no matter how good they are at writing, don’t understand what they are writing, and can’t develop new ideas, only combine existing ideas together. Well I guess not everyone agrees with that, because I don't for example. First, how different are the two things, "develop new ideas" vs "combine existing ideas together"? Are we certain that some things that you consider new ideas aren't ever things I would consider the combining of existing ideas? Regardless, it seems to me that these language models may very well produce "new ideas" by chance, even if it doesn't itself 'recognize' that that is what it's doing. > Given that, it’s seems that what you are doing adds only noise, and no value, to HN. I don't agree that the only value added to HN is 'new ideas'. There are lots of old ideas that have lots of value in being communicated and discussed.