Hacker News Reader

Comment by famahar | Hacker News Reader

I uploaded my detailed game design document for a project I've been working on in my free time and it was kind of a weird confidence boost. The two hosts seem to treat ideas like their the most insightful relevatory information they've ever heard. After a few uploads of other documents you start to notice the same overly surprised tone.

View on HN

The prompt affects that a lot. If I input my writing and ask an LLM to “evaluate,” it will tell me how astute and intriguing my ideas are (often to the point of hyperbole). If I ask instead for it to “critique” me, I’ll get a much less complimentary response about the same content.

You don't get prompt control here, just one button.

There is limited prompt control. You can provide podcast production notes for the AI to play off of, and they tend to honor it. I don’t know that they’d so completely change their personalities to suit it though.

Must have been prompted to be an American podcaster.

Bring on the one that's all British and snarky!

Might be hard to spot if you’re American in the US, but LLMs feel very American even outside of region and language. Concretely, I have to constantly ask for recipes to be changed to metric. Less concretely, the undertones and mannerisms of politeness, positivity and “excitedness” comes across as very American to me, probably even within the rest of the Anglosphere. How would I describe it? Maybe similar to how you’d feel about a mix of Ned Flanders, Ted Lasso and some Valley girl stereotype – im sure it’s a bit off putting also for many Americans.

I guess it’s training data but also heavily RLHF. I doubt that the trainers are aware of their own cultural biases and values, and they may not care. And why should they? In either case, from a thousand yard perspective, it’s probably an effective vector for spreading “American values”, if you will.

Oh absolutely. Early on into the AI craze I tried to use it to summarise my messages[1] and it made them overly fluffy and weird.

Anyone receiving the message would instantly clock that I didn't write it - even with a prompt longer than the original message trying to massage out all of the Americanisms and false enthusiasm. Not a use case that works for me, haha.

[1]: I was trying to use it to shorten my "If I had more time, I would have written a shorter letter" waffling.

I guess there might be some culture nuance here.

I can't speak to all of the LLMs, but as an American who listens to a LOT of podcasts, I can tell you why these ones sound the way they do: the audience. People who listen to (non-fiction) podcast want to be informed. They are people who are curious about the world around them and are generally interested in self-improvement at some level. Can you imagine a personal finance or health podcast delivered in a pessimistic or even fatalist tone? No, they are all _optimistic_ (even energetic) in tone, because that's the WHOLE reason people are listening to them at all.

I don't think the folks at Google are as patriotic as you think they are.

As an American, I find it exhausting. I think of it as fake Silicon Valley/SF "nice" affect combined with non-US English as a second language floridity. My setup prompt for ChatGPT includes a reminder that if it answers too long or slowly, then it will take time away from my medical research grad students and people will probably die as a result of the delays. It helps a little.

This seems to be a common trait of a lot of the more "aligned", "helpful" LLMs out there. You can drop any random excerpt from your diary into ChatGPT and it will tell you about how brilliant, sensitive, and witty you are. It's really quite sickening.

Reminds me of my father who'd tell every kid that they're a genius, including myself. It got me motivated to try things, but whenever there was a failure, I felt terribly betrayed.

General advice from psychology is that when it comes to success you should praise the kids for things they control, like effort, time spent, inquisitiveness, concentration not things that are out of their control like talent or luck. Basically praise for what they did, not what they are.

When it comes to morality, it's the other way around. You praise kids for being good people when they do something right. Because you want them to internalize identity of a good person and associate it with those behaviors.

Internalizing identity of a genius is mostly useless, rarely beneficial, often harmful.

That sucks. But it's why I keep trying to remind my kinds that even though they are smart, they will fail at things. Failing is a part of learning. Possibly even the most important part. "If you're not making mistakes, you're not trying hard enough."

Honestly, it's obviously horrendously gag-worthy and everything, but also kind of funny that there is so much bullshit marketing copy out there that LLM's invariably converge on this inspirational Stanford application letter / upbeat linkedin influencer tone of voice, and just apply it to everything.

Well, an LLM doesn’t have the capability to like anything more than anything else. It doesn’t really matter to GPT if your diary excerpt is the worst piece of writing ever written, or the most brilliant - it’ll just tell you what you want to hear and that’s that.

Only because they've been RLHFed and prompted to be agreeable. A Marvin the Paranoid Android LLM could similarly be designed to hate everything equally.

Genuine People Personalities, indeed.

"Tell you what you want to hear" is a matter of training and prompting, not the technology itself. But I agree that asking an LLM to make an aesthetic judgment is a fool's errand.

How is it sickening? Tell it to roast you if you think it's a problem.

It feels sickening to be praised meaninglessly for something not worthy of praise. ChatGPT in particular loves to talk about how clever and interesting text you show it is, even if you're not actually asking for that kind of analysis.

It's also sickening that I see people using these LLMs to rewrite performance reviews, peer feedback, business reports, etc. I've already started to notice business communication getting even more saccharine and toothless.

Sickening in the same way you get sick from eating too much sugar.

I was suprised at how effective the positivity was on me when I fed it one of my design docs! Color me impressed at the naturalness of the resulting "conversation".

Reply on news.ycombinator.com