Hacker News Reader

Hacker News

Comment by throwaway888abc | Hacker News Reader

Parent Full thread

throwaway888abc

You can create counterpoint arguments with ChatGPT...

Sht in / Sht out. Everybody happy.

Years back during big outsourcing project, this never ending conversation (between human teams) were big part of it.

View on HN

I'd say tell them to ask GPT why it won't work. It'll say something like "You're right, I've made a mistake...".

This should illustrate that it can be helpful, but it's just aiming to please whether or not the solution is workable or sane.

Just asking "Does that work?" can demonstrate this. Slightly contrived example (I've trimmed the ChatGPT output down - it gives some warnings about code readability and that it only has knowledge up to 2022 that might be useful if the code worked at all):

> User: how can I pass two parameters to a hugo partial as positional parameters (without using a dict)?

> ChatGPT: If you want to pass parameters as positional arguments without using a dictionary, you might need to work around this limitation. One approach is to use a custom data structure, such as a slice or an array, to emulate positional parameters. Here's an example:

    {{/* In your main template */}}
    {{ partial "yourPartial.html" "value1" "value2" }}
    
    {{/* In yourPartial.html */}}
    {{ $param1 := index . 0 }}
    {{ $param2 := index . 1 }}

[ChatGPT has also suggested in other runs that the values can be accessed with {{ .Param1 }} and {{ .Param2 }}]

> User: Does that work?

> ChatGPT: I apologize for the confusion in my previous response. I made an error in the example provided. Hugo's templating system doesn't support passing parameters as positional arguments directly, and my previous example was incorrect.

Have you tried asking it if it works when it gives a correct response? :)

Indeed, ChatGPT will frequently backpedal if you even slightly hint that its answer may have been wrong: https://arxiv.org/pdf/2305.13160.pdf

I have found that you can reduce that behavior if you prompt it correctly, though.

Is this really your suggestion to someone dealing with a coworker making your work harder to do by gumming up the works with a bad tool? This is cult like behavior.

If the coworkers are convinced ChatGPT is a solid source of truth, showing them that ChatGPT will prove their generated answer wrong sounds more effective and time efficient to me than handwriting a refutal that they won't trust as much as GPT.

I don't see the problem. What's the issue? What is cult like about this?

Or you can ask ChatGPT for the most insulting comment it can come up with. When they get upset, tell them ChatGPT generated it. Look surprised, like you just had the idea, and say that maybe ChatGPT-generated content in emails isn't actually a good idea.

You can type shit here. It's ok, we all know what you meant anyway.

Reply on news.ycombinator.com