you are viewing a single comment's thread
view the rest of the comments
[–] 15 points 2 days ago (2 children)

AI has never been 100% accurate to the documents it pretends to reference, so the fact that anybody could even consider reading it like it's not pure gibberish is mind boggling to me.

  • source
  • hideshow 2 child comments
  • [–] 7 points 2 days ago (1 child)

    I gave Claude a document to proofread today. It came back with an error. Clause said I used "told" instead of "tell." The error ended up being Claude's error, as I used "tell," and it was just confused.

    This was the first time I ever used Claude. Just wanted to see what it was like. It did apologize, in fairness. But yeah, it's clearly fallible. I guess I don't understand it well enough to get how it read the document incorrectly. I'm baffled.

  • source
  • parent
  • hideshow 1 child comment
  • [–] 5 points 2 days ago

    It's because it's a statistical model that guesses the next word. You asked it to read a document and formulate a response. So, based on responses in the dataset to the documents in the dataset, it chose the words and phrases which were most likely to be in a response.

    Statistically, the most likely result the machine could come up with, was that you had used the wrong word. Factually, you did not.

  • source
  • parent