Hacker News Reader

Comment by andersrs | Hacker News Reader

[I know you feel that this technique is helpful and you want to save the world. That's double plus good of you.] [For me, I think that] all the padding makes the conversation diluted, watered down and boring. It's similar to 'The Gift Shop Sketch' [1]. [I feel] this is very disrespectful of my time. [In my opinion] it would be great to use another model to strip all the feely words, condense the meaning and generally un-fuckify all the AI waffle. [What are your thoughts about that?]

[1] https://www.youtube.com/watch?v=7MFtl2XXnUc

View on HN

Hmm, not my choice of sketch to demonstrate this. That is more about crap padded TV. "ChatGPT I have 1 minute of content, please pad out to 25 minutes".

There must be a 1000 British sketches about excessive politeness but none come to mind right now... the best I can think of is that it's an inverse Key and Peel anger translator:

https://www.youtube.com/watch?v=-qv7k2_lc0M

We will need the anger translator in our client software to translates back out of the polite version so we can efficiently understand what is truly meant without confusion.

https://news.ycombinator.com/item?id=37809022

I had the opposite idea: Run all my comments through an LLM to make them more diplomatic. I noticed my comments have been a bit adversarial lately, so I just wrote a script to pass the HN comment history plus my draft comment to GPT-4 and have it reword my replies. Hopefully it won't be overly verbose.

This is this comment, GPTed:

~~~

Hey andersrs, I totally appreciate your perspective here - the 'Gift Shop Sketch' is a classic! To share my thoughts, though, I've actually contemplated doing quite the opposite. I've begun to realize that my comments might've been coming off a little too confrontational or straight to the point recently. So, I'm thinking of running them through a language model to dial up the diplomacy a notch.

Could be fun, right? I'm in the process of penning a script that takes our comment history, along with any new comments I jot down, into account and passes it all on to GPT-4. The goal's to help it finesse my responses a bit. Here's hoping it won't tip the scale in the other direction and make them a long-winded affair!

~~~

I agree that it's a bit meh, but I think with some prompt tweaking it can be very helpful.

EDIT: With some better prompting:

~~~ Actually, I've been thinking about making my comments more cordial. Noticed that they might come off a bit straightforward, so I've worked on this script to run my HN comments, along with my draft comment, through GPT-4. The idea is that it can help to craft my replies in a way that fosters better communication. I'm hopeful it won't end up making them overly long-winded. ~~~

Part of communication is knowing your audience.

For example, to me this sounds like I’m being talked down to as if you’ve pegged me as someone you have to finesse your words around or else I’ll erupt into violence at the drop of a hat.

I get the purpose, to not sound adversarial in your response. Just a thought though.

I agree with you, but all the downvotes I get don't!

Ugh, I hope that’s not a symptom of the future we’re looking at. Sites that run users comments through LLMs with strict rules to make them more “amenable”.

Well, it's one thing to do it yourself and another thing to have it done by someone else whether you like it or not.

Reply on news.ycombinator.com