pull down to refresh

50 sats \ 0 replies \ @kepford 4h
When you learn how it works.. these mistakes make sense.
reply
Fake news.... I just asked it "how many b's are in the word blueberry" and it answered 2.
reply
0 sats \ 1 reply \ @carter OP 4h
I just got it to respond incorrectly to how many r in congratulations https://chatgpt.com/share/689a478b-e6ec-8008-b45b-4fe285cb27a2 even when I said to check with python
This is old explanation of why it gets it wrong. https://www.reddit.com/r/OpenAI/comments/1haxhjk/can_someone_explain_exactly_why_llms_fail_at/ I also noticed it seems to do better if it "thinks"
reply
reply
If it depends on a specific form of imputation to respond correctly, it is because it continues to be what it always was, a calculation presented in the form of language and not something truly intelligent.
reply
The criticism is extremely valid, since people forget that, like any software or machine that depends on human input, form matters. It is worth remembering that, unlike marketing, the word “intelligence” is not something that is present in these things.
reply