Link: “ChatGPT is a Blurry JPEG of the Web

Original post found at: https://www.newyorker.com/tech/annals-of-technology/chatgpt-is-a-blurry-jpeg-of-the-web

Interesting article talking about how “artificial intelligence” like what ChatGPT uses is basically comparable to a lossy compression algorithm. It hoovers up information from the web but doesn’t save all of it, and its text-generating capabilities (as in, its “given a certain prompt, what’s most likely to be the first word, and then what’s most likely after that, and after that…”) obscure what it’s really doing. It’s like a search engine that regurgitates some predictive-text-generated spiel instead of just showing you the original result.

I also appreciated the detail that ChatGPT is not capable of basic addition or subtraction, unless it saw the answer on one of its scraped web pages. Apparently it hasn’t learned how to carry the 1. lol