Hacker News Reader

Comment by altdataseller | Hacker News Reader

While I’ve had SO mods criticize my questions, I’ve more success getting correct answers to questions from SO than ChatGPT. With SO, I also have the benefit of knowing answers have been mostly vetted by others. If it’s wrong, someone will comment on it usually. Not so with ChatGPT

View on HN

It just depends on the questions. I have gone from asking 2 or 3 questions per months (and reading 100s per months) to using ChatGPT for all of it. Because it gets the questions better and the response is immediate.

GPT 3.5 or 4?

You know, I haven't heard anybody talk about this yet! I was wondering, is gpt4 better than 3.5, or worse?

It's miles ahead of 3.5, the difference is really really big. Mainly because GPT-4 is just so much better at "understanding" and logic, it follows everything you ask for or say very, very closely.

Incredible! I am sure its amazing to use. That said, I doubt it could, say, answer questions from an LSAT exam and get them right though. That would just be way too unbelievable to me.

I'm not sure if you're joking or not, because OpenAI actually used LSAT as one of the examples where GPT-4 is much better - https://openai.com/research/gpt-4

https://files.catbox.moe/l5n5gd.jpg for some of the results in their technical report

Amazing! How lucky we are to be alive for such advancements in Technology. They are really doing something special at OpenAI.

Chatgpt4 is better than 3.5, but I think it’s also because 4 uses larger models because you are paying, while 3.5 has been cut down. I still find that 4 will give bad technical answers … for example, ask it to help you write a Caddy reverse proxy config and it will give you garbage very emphatically

Gotcha. I usually use nginx anyway, so no worries there.

Much better

Wow!

Reply on news.ycombinator.com