Hacker News Reader

Comment by thereddaikon | Hacker News Reader

I hardly use it at all. ChatGPT was interesting at first but it slowly got worse and worse as changes had to be made to it so OpenAI wouldn't get sued. The same is true for all the other ones. LLMs can be great if they are allowed to blatantly violate copyright. But once everyone started circling the wagons and walling off their information they quickly started to lose their value.

View on HN

Feel free to not use it, it's your time.

Some recent examples:

* "I installed windows and now I can't access my linux partition" * "I want to add an `in` variable to an existing pl/sql function, can it go on the end of the parameter list? (Answer: No, because...)" * "What size soccer ball does AYSO U10 use?" * "what is the name of that python library that handles command line arguments, and could you provide an example that demonstrates required flags?"

nothing earth shaking. not things that're impossible to find, but... instead of reading someone's blog post in which they work as hard as possible to belabor the point so you'll scroll through all 14... this thing just gives you the answer.

The last time I used an LLM I asked it to modify a splunk query for me. The result wouldn't work so I gave it feedback. It was still wrong. I tried coaching it a few times but it couldn't make it work and by that point I had spent more time telling the LLM how to write a splunk query than I would have spent doing it myself.

Yep, I had a similar thing happen. I asked it to give me a set of regexes and after spending a couple minutes trying to correct it, I just wrote the script myself. It's neither useless nor the all-seeing eye. Instead, it's just a tool.

Reply on news.ycombinator.com