News for hackers. Stuff that matters.

A comment on ChatGPT Gets Its “Wolfram Superpowers”. Parent comment

Re:ChatGPT Gets Its “Wolfram Superpowers” by danielmarkbruce on Thursday March 23, 2023 @06:13PM (#35282058)
Imo there is less of a flywheel than it might seem. It's not like google search where a click of the 7th result is concrete feedback. ChatGPT doesn't get concrete feedback (it's not none), and the amount of text available for training is crazy high.

Additionally, the model doesn't learn from experience. If you use GPT-4 for 10 years, it will behave the same in the 10th year as the first. It's not getting better as you use it. OpenAI could improve their models based on the feedback they get but I believe they claim they aren't doing this, and I doubt they'd lie about that.

[ Parent | on HN ]
Re:ChatGPT Gets Its “Wolfram Superpowers” by SomewhatLikely on Thursday March 23, 2023 @07:17PM (#35282868)
Sounds like you don't ever submit feedback. It's been there from day one and asks you to write why you didn't like the response. It would not shock me if they use gpt itself to read the feedback and provide a corrected response for the next versions training data. They could also log anytime some provides negative sentiment in the chat session for training signal.
[ Parent | on HN ]
Re:ChatGPT Gets Its “Wolfram Superpowers” by danielmarkbruce on Thursday March 23, 2023 @10:39PM (#35284389)
Hence the "it's not none" part of the comment. There is nuance...
[ Parent | on HN ]