4 ms·
When my brain generates the next wurd I'm perfectly capable of taking decisions of misspelling "word" for "wurd", LLMs can't make such reasonings unless instruc
by epolanski 3y ago
When my brain generates the next wurd I'm perfectly capable of taking decisions of misspelling "word" for "wurd", LLMs can't make such reasonings unless instructed to act like that.
- david-gpu 3y agoWhy would I want an AI assistant to have agency? I want them to help me, not to further their personal goals. In fact, I don't want them to have personal goals other than helping people.
- epolanski 3y agoI didn't say it should have one, I'm saying that LLMs statistically finding the next bit of information aren't really making decisions, which is to counter-argue the fact that it's not different from how we reason.
- david-gpu 3y agoThank you, I had misunderstood your point.
- HDThoreaun 3y agoYour comment doesn't convince me you're making decisions either. "Wurd" could've just what you considered the best token to get your point across in the same way that LLMs choose the best token