So one of my students, she had a problem and she was asking um and she asked good questions and rather than give her the answer directly, I was training her to ask better questions because then she could sleuth and investigate her way to the answer. Yes, it was longer, it was more painful, it took a lot more time for me and it took a lot more time for her, but she was learning critical skills in the process. And then eventually she said, "Oh, I'm just going to ask ChatGPT." Um and she asked ChatGPT and she came up with the answer and she sent it to me and I said, "Okay, looks all right. Uh I'm not going to say anything." And I gave her a prompt, right? Quincy, I gave her the prompt and the prompt is, "This is the answer you gave me last time to this and this question. Um Now, pretend you're a very senior software engineer and tell me why this is wrong." And ChatGPT told her in detail why she was wrong why its own answer was wrong. And she came back to me and she said, "This has just blown my mind. I've I don't know what's right anymore." And I said, "This is always the problem is even the expert that you get advice from, you don't know if they could be wrong in your circumstance, right? And ChatGPT doesn't know anything. It's calculating the probability of every next word. It doesn't know the answer. It's calculating the probability of words in sequence. Okay, so don't assume it's omniscient. It doesn't know everything. It's just really good at calculating probability. Now, it so happens its answer was directionally correct, useful, a good stepping stone, not entirely wrong, but it then went and said its answer was wrong, right? Now, who do you believe? So and that's why I told I said you have to think about this from first principles. >> wants to do whatever you told it to do. So it'll figure out reasons why what it previously said is invalid because it's trying to make you happy at the end of the day. >> 100%
Generated algorithmically for Search Engine
Indexing.