Comment by AstroBen

I had a weird LLM use instance happen at work this week, we were in a big important protocol review meeting with 35 remote people and someone asks how long IUDs begin to take effect in patients. I put it in ChatGPT for my own reference and read the answer in my head but didn't say anything (I'm ops, I just row the boat and let the docs steer the ship). Anyone this bigwig Oxford/Johns Hopkins cardiologist who we pay $600k a year pipes up in the meeting and her answer is VERBATIM reading off the ChatGPT language word for word. All she did was ask it the answer and repeat what it said! Anyway it kinda made me sad that all this big fancy doctor is doing is spitting out lazy default ChatGPT answers to guide our research :( Also everyone else in the meeting was so impressed with her, "wow Dr. so and so thank you so much for this helpful update!" etc. :-/

If the writer’s entire process is giving a language model a few bullet points… I’d rather them skip the LLM and just give me the bullet points. If there’s that little intent and thought behind the writing, why would I put more thought into reading it than they did to produce it?

And what's more is the suspicion of it being written by AI causes you to view any writing in a less charitable fashion. And because it's been approached from that angle, it's hard to move the mental frame to being open of the writing. Even untinged writings are infected by smell of LLMs.

Here's my take - art has value because of the context it is created in. The author's history, current events that we live through as groups, the reactions to a work being released, availability of materials - all these things are fundamentally human. I believe the reason art has value to us is because of the empathy and humanity that we all share despite major differences in beliefs.

That's not to say computers can't generate beautiful things, but unless you expand the context out to include the history of how a program that can create such art came to be, the output is not meaningful. This is why people do not react well to AI art made from simply throwing prompts at a model, or writing that does not feel like it has style, struggle, or any personal flavor.

I've always believed that LLMs will be able to fake it perfectly one day. But as a music fan, no fully computer-generated music will ever bring me the range of emotion and joy that another human's story and creative process through that story does.

A person can be just as wrong as an LLM, but unless they're being purposefully misleading, or sleep-writing, you know they reviewed what they wrote for their best guess at accuracy.

Writing nice sounding text used to require effort and attention to detail. This is no longer the case and this very useful heuristic has been completely obliterated by LLMs.

For me personally, this means that I read less on the internet and more pre-LLM books. It's a sad development nevertheless.

Art, writing, and communication is about humans connecting with each other and trying to come to mutual understanding. Exploring the human condition. If I’m engaging with an AI instead of a person, is there a point?

There’s an argument that the creator is just using AI as a tool to achieve their vision. I do not think that’s how people using AI are actually engaging with it at scale, nor is it the desired end state of people pushing AI. To put it bluntly, I think it’s cope. It’s how I try to use AI in my work but it’s not how I see people around me using it, and you don’t get the miracle results boosters proclaim from the rooftop if you use it that way.

You're absolutely right! Art is the soul of humanity and without it our existence is pointless. Would you like me to generate some poetry for you, human?

Agreed, except s/know/think. It's possible that there are some false positives in my detection algorithm, that I tune out just because someone's prose style has that undercurrent of blandness characteristic of LLMs. But I suppose if we're talking about "art" and not, for example, technical documentation, that's no great loss --- bland writing isn't worth recreationally reading.

It does seem that LLMs could avoid this detection with some superficial tweaks such as injecting poor grammar and reducing peppiness. I hope it doesn't get to the point that I have to become suspicious of all text.

I couldn't agree more. I don't care how much prompt or model finetuning you did, if something is shat out by an LLM I'm not interested even remotely.

I wish more people held the same opinion actually. Unfortunately, my sense is that most people don't care, they are fine with LLM generated crap

It honestly makes me want to blow my brains out

If you think about how an LLM works, it’s rounds off the outliers in its training data so the result is sort of averaged and homogenized. Art and writing are an expression of the very thing that LLMs discard - our unique qualities, our outlying quirks, that make us more than just another human.

> There's something unique about art and writing where we just don't want to see computers do it

Speak for yourself. Some of the most fascinating poetry I have seen was produced by GPT-3. That is to say, there was a short time period when it was genuinely thought-provoking, and it has since passed. In the age of "alignment," what you get with commerical offerings is dog shite... But this is more a statement on American labs (and to a similar extent, the Chinese whom have followed) than on "computers" in the first place. Personally, I'm looking forward to the age of computational literature, where authors like me would be empowered to engineer whole worlds, inhabited by characters ACTUALLY living in the computer. (With added option of the reader playing one of the parts.) This will radically change how we think about textual form, and I cannot wait for compute to do so.

Re: modern-day slop, well, the slop is us.

Denial of this comes from a place of ignorance; let the blinkers off and you might learn something! Slop will eventually pass, but we will remain. This is the far scarier proposition.

I would love to see true really good AI art. Right now the issue is that AI is not there where it by itself could produce actually good art. If we had to define art it would be kind of opposite of what LLMs produce right now. LLMs try to produce the statistical norm, while art is more so about producing something out of the norm. LLMs/AI right now if it wants to try to produce out of norm things, it will only produce something random without connections.

Art is something out of the norm, and it should make some sense at some clever level.

But if there was AI that truly could do that, I would love to see it, and would love to see even more of it.

It can be clearly seen, if you try to ask AI to make original jokes. These usually aren't too good, if they are good it's because they were randomly lucky somehow. It is able to come up with related analogies for the jokes, but this is just simple pattern matching of what is similar to the other thing, not insightful and clever observation.