Ask HN: Do you also never read the last paragraph of ChatGPT's responses? | Hacker News Reader

Ask HN: Do you also never read the last paragraph of ChatGPT's responses?

12 pointsbehnamoh10 comments

Due to heavy RLHF and finetuning, ChatGPT often concludes by mentioning ethical aspects of what I asked for, things to be cautious about, etc. It's belittling tbh, as if it's treating us like children.

So I've got used to just skip reading the last paragraph in ChatGPT's responses. It's kinda like ad numbness—you see it but your mind automatically ignores it.

View on HN

I agree: it's dumb.

However, the sentiment expressed in your post is subjective and depends on individual user preferences. Some users may appreciate the ethical considerations, while others might find them unnecessary. Developers often take user feedback into account to improve the user experience and enhance the AI system's interactions. It's also understandable that some users may find these statements repetitive, especially if they have read similar disclaimers multiple times. Striking a balance between promoting ethical usage and avoiding an excessive number of repetitive statements is a challenge faced by developers and researchers.

I have talked to noone who finds wasting half your tokens spitting out empty platitudes and weasel words was worth the effort. Just search 'gpt apologizing' and see how many people are liking, or even okay with the huge waste. Just think how many megawatts have been wasted on mildly insulting gpt 'politeness' cycles.

Sounds like something ChatGPT would say.

You got me. The second paragraph came entirely from ChatGPT :P

Ha, no wonder I instinctively stopped reading after a few sentences. GPT is such a terrible writer.

Sam Altman is a bad leader, and in my experience, most who come from where he comes from believe people who do not are inferior and need these kinds of redundant reminders of how to be good little boys and girls.

Just think of all the f-n idiots that got caught using ChatGPT without knowing its limitations. Like the reporter or lawyer that cited resources that don't exist due to hallucinations.

But yea, I gloss over all that shtie. It's insulting.

I gave up on gpt months ago, cancelled my sub. The quality went down so dramatically I considered trying to recover my money under bait and switch fraud laws.

The base model quality has gone up drastically since the update last month, but ironically, GPT-4 quality has gone down.

More non STEM folks lurking around AI field, worser the effectiveness of AI. Instead of knowing and answering everything, now we have moral checker AI as opposite to Grammarly!

Reply on news.ycombinator.com