The Core

Long indeed. I didn't read but a portion of those 17,000 words (possibly more because it was long before the end that he said he was already at 17,000 words).

I am not surprised by his findings in general, but he did change my mind about how to think about ChatGPT hallucinations. If ChatGPT is supposed to give authoritative answers that can be trusted to make decisions, the hallucinations are a problem. But if we are just comparing it to the "hallucination-adjacent" answers you get from Google and Bing, then actually it starts to look pretty good.

And, Brad, did you swoon with love and appreciation when you saw that home page? I don't think it's possible to go more minimalist than that.

And I loved this quote from Paul Graham on one of the other posts: "What nerds like is the kind of town where people walk around smiling. This excludes LA, where no one walks at all, and also New York, where people walk, but not smiling." I had to hunt for the original and find it here

And ChatGPT can't distinguish Truth any better than Google can.

The bold indicates the reframing that changed my mind. You always hear about how ChatGPT is kinda junk because it hallucinates. But his reframing of saying that it should be evaluated compared to Google, not compared to a research scholar in the domain in question, is interesting. It underlines the fact that the morass of junk returned by Google is indeed a morass of junk, especially in highly commercialized spaces.