3 ms·
I’ve seen the same behavior. I wonder if these are learning instances for the AI or is it just dumbly going through a set of possible answers until the user is
by kmos17 4y ago
I’ve seen the same behavior. I wonder if these are learning instances for the AI or is it just dumbly going through a set of possible answers until the user is satisfied with the answer.
If it’s the latter, it’s basically a fancy way of doing a google search and going through potential answers.
If it’s the former and the AI learns from these mistakes (say if a consistent number of users point them out) I assume it will get much better over time.
- kmos17 4y agoanswering my own question, I guess not: https://help.openai.com/en/articles/6787051-does-chatgpt-remember-what-happened-earlier-in-the-conversation https://help.openai.com/en/articles/6787051-does-chatgpt-rem... So at its current stage it’s pretty useless as it will not learn from its mistakes, seems to me the opposite of intelligence, just another mechanical turk.
- _db4z 4y agoI think the two main problems here are: - Users can't use it to learn, because ChatGPT responds with high confidence but could be terribly wrong. - Even if it could learn from users, it would really only learn from people that know a subject and are willing to spend their time correcting ChatGPT.