For me the biggest problem is the lack of attention to the output itself. It's great to use AI as an amplifier and enhancer. And where it really can do the job itself, excellent! But plenty of people publish AI content without so much as reading or listening to it before doing so, and that I find really frustrating.
Recent example: My wife came across a guided meditation where, about 5 minutes in, they said to picture a green light in your chest, 'this is your heart slash custom phonemes zero slash.' Obviously someone incorrectly formatted a file or variable name for the word 'chakra' and their AI voice read it as words. Clearly no human even listened before hitting publish.
Actually, I told one lie at the beginning.
This essay was one-shot by Claude.
How do you feel about the opinions espoused within? Are they less valid?
.
.
.
Nah, just kidding. I've learned my lesson, it's all Daniel.
That torpedos the whole essay just as much as if it's AI slop, and for the same reason. You've deliberately torn up the contract with the reader. I no longer care which part of this is true, because either way, I no longer care about the essay.
I hear your point, but I disagree on the contract.
The quality of the message should be inherent. I'm nudging readers to consider whether they value the essay less if they think it's AI.
Clearly it goes from having potential to being trash. But the content didn't change. That should give pause, no?
I put my name on this work and vouch that it's worth your time. Yes. Why does it matter how it was made?
The quality of the message should be inherent. I'm nudging readers to consider whether they value the essay less if they think it's AI.
The message is not contained in the words, but in the words together with their generating process [1] — how they came to be written. [2]
I'm nudging readers to consider whether they value the essay less if they think it's AI.
I do value a message less, often as low as zero, if it was generated by an AI, and need no nudging to have made that decision.
Clearly it goes from having potential to being trash. But the content didn't change. That should give pause, no?
The only thing that did not change was the words.
The words of your best friend, an experienced car mechanic, might not differ from what the used car saleman tells you about the car you are looking at, but you will trust one over the other. The words mean different things coming from different sources.
Consider the Open AI/HuggingFace debacle, which gets worse the more we find out. Would you trust what these AIs told you?
I put my name on this work and vouch that it's worth your time. Yes. Why does it matter how it was made?
With respect, as you're a new account here, I know nothing about you and do not know how much that assurance is worth — how much assurance it gives me about the generating process, if you did use AI in some way. [3]
This applies to all purported evidence of anything: what was the generating process? What process created it and brought it to my attention? Many paradoxes in probability and statistics can be dissolved by asking this question. ↩︎
Borges imagines an extreme example in his "Pierre Menard, author of the Quixote":
It is a revelation to compare Menard’s Don Quixote with Cervantes’. The latter, for example, wrote (part one, chapter nine):
". . . truth, whose mother is history, rival of time, depository of deeds, witness of the past, exemplar and adviser to the present, and the future’s counselor."
Written in the seventeenth century, written by the “lay genius” Cervantes, this enumeration is a mere rhetorical praise of history. Menard, on the other hand, writes:
". . . truth, whose mother is history, rival of time, depository of deeds, witness of the past, exemplar and adviser to the present, and the future’s counselor."
History, the mother of truth: the idea is astounding. Menard, a contemporary of William James, does not define history as an inquiry into reality but as its origin. Historical truth, for him, is not what has happened; it is what we judge to have happened. The final phrases – exemplar and adviser to the present, and the future’s counselor – are brazenly pragmatic. ↩︎So I did some due diligence. Is this you? I liked the Chernocubes. If you're in London, as well as working there, have you seen the Escher exhibition at Somerset House? ↩︎
I completely agree that whoever sent the message is crucial metadata. And therefore, it's kind of refreshing and powerful to see people who use their real name on the web.[1]
The only thing that did not change was the words.
That's a powerful point. I guess I'm just partial to the idea of the inherent value of an idea. Isn't that what motivates you to do science[2]? That you can leave behind results that stand on their own merits, even if your name and reputation are forgotten?
For me, it's primarily two things:
- I really dislike dishonesty, and publishing AI generated content as your own seems quite deceptive. I'm much less likely to have a bad opinion of an AI piece of any kind when the human behind it is upfront about how it was made. Of course, not every use of AI necessarily has to be disclosed and it's a fair question where the boundary should be. But when, say, someone posts a post on lesswrong where the majority of words are LLM-written, or publishes a song made with Suno - however much effort the human has put in - that imho deserves disclosure.
- The basic question about distributions, priors, and how I can avoid wasting my time. I'd assume that 99% of clearly AI-created media (whether Suno songs or texts or shorts of inspiring old people in armchairs) really is low effort slop and I'm well advised to skip them as soon as I realize that. Sure, this comes at the cost of also skipping the 1% that do include deliberate effort and intention. But if the alternative is that I drown in slop just because maybe that instance is different, then I'm happy with being very occasionally wrong in my slop assessment. Notably, I likely won't skip such cases if the creator is open about how it was made and perhaps even points out while it might still be worth my time.
I don't think the OP was hiding the parts of the video that were created with AI though.
Consider the Impossible Burger. If you give someone "a burger" (without being specific about its makeup) and they later find out it was Impossible Burger instead of ground beef, some people will be impressed with the quality of Impossible Burger, but some people will be angry that they were deceived. The angry people will be a lot louder than the impressed people. And the fact that you successfully fooled them makes it worse!
Yeah, the song has the generic AI quality, I couldn't get through more than 10 seconds.
I wouldn't make big conclusions about what people like or dislike. Maybe AI will learn to make good art, and maybe it'll even happen soon. But it hasn't happened yet. Today's AI is much better at math/programming than at drawing/writing/music, and it's best not to use it for anything creative.
In fact, since the generic AI quality of drawing/writing/music is so often easily noticeable, I think it must be machine noticeable as well, and I'm super confused why labs don't just train it out. If I didn't hate everything about AI, I'd probably make a startup on this idea and make a ton of money.
Your stance is what the post is about.
Pre-AI people used a set of heuristics to verify whether a piece of art was worth their attention. These heuristics included meeting a minimum technical bar, and being catchy. Most of the people evaluating songs do not have experience writing music, hence cannot judge on anything deeper.
We trained AI on things that matched the good output quality heuristics. These collapses into a few distinct surface styles.
Because of the initial problem (people lacking the experience to judge output), those creating with AI go along with the default surface styles, still using the pre-AI heuristics. And those judging for slop will apply their new slop detection heuristic to ignore anything with the default surface style, even if a lot of work was put into some other part of it, like the lyrics.
(I do hate a lot of the AI content I prompt, now, too, but at least it's instrumentally useful for me.)
For me, I find this varies hugely by which model I'm talking to. Claude Opus 4.6 was a bit annoying but bearable. Opus 4.8 and 5 are just agonizing—they use one of their cliche phrases once every 2-3 sentences. And Opus 5 is nearly allergic to concrete names when a vague reference could be used instead.
But a model like Qwen3.8 27B, while it has picked up a few of Opus's favorite bits of vocabulary, limits itself to using "load bearing" once every couple of pages at most. And it generally just seems more pleasant, while still being willing to flag potential design issues for me.
So I think about half my frustration when talking to AIs is simply because recent Opus releases are terrible writers. [1] The other half may be more general.
I have had other frustrations with OpenAI's models dating to earlier this year, but that's a whole other topic. And Gemini is scarcely worth it in terms of price/performance these days. ↩︎
I bounced off this - seemed long and rambling, and it was unclear what the thesis or point is. I only came back when a commenter mentioned the coda.
Actually, I told one lie at the beginning.
This essay was one-shot by Claude.
How do you feel about the opinions espoused within? Are they less valid?
No way was it one-shot, unless it was a monster prompt or seeded by a fair bit of other writing. I couldn't easily identify opinions (just rhetorical-ish questions and not-quite-relevant statements that don't answer the question), so they're no more nor less valid. I don't think the examples describe me or the people I know well, so maybe that interfered with my inferring opinions from them.
I think AI would have made it much more top-down and to-the-point.
The one line takeaway is: People are using the proxy "looks like AI" to mean "not useful" and I think that's wasteful.
Half-agreed on both points. AI would have improved this, but perhaps not above the threshold of being worth the time of most LW readers. With some AI AND human effort to explore what points are novel or contested before writing about half the length of post, it could be quite good.
People are using "looks like AI" to mean "it's not useful", but I think that's efficient, not wasteful. Taking the human effort to critique and specify WHY it's not useful would be wasteful.
Well it's my first post.
Do you mean go argument by argument / example by example, and really stress test each one and only keep the best?
Hope to improve, so thanks for the feedback!
Welcome, and I hope you'll refine and post further - I didn't downvote, it wasn't actively bad, just kind of ... meandering. The autobiographical parts likely do resonate with some readers, but I'd prefer a more direct thesis followed by the anecdotes and evidence.
I also hope you won't make the mistake with my comments that this post seems to make, of over-generalizing from a few experiences. I'm probably not the median LW reader or commenter, and both LW and Reddit (and specific subreddits) have a pretty wide range of preferences, so "they HATE" or "X is important" generalizations should be limited a bit ("the most vocal on X HATE" or "X is important to many consumers of art").
This essay was one-shot by Claude. How do you feel about the opinions espoused within? Are they less valid?
My reaction was to be impressed by your prompting skill, with a (unendorsed) flicker of "yeah that makes sense, this was really high quality with insightful ideas, of course AI was involved". But I live and breathe AI, so, not your target audience.
(This essay was written entirely by a human)
I was banned from Reddit recently. This after being a constructive member of the community for almost a decade. If that’s not an oxymoron.
I was quite baffled. Let me explain.
My stint as a bard
About a decade ago, I began work on a satirical song. At the time, I had practically none of the skills needed to produce music. What I had was an idea: I was inspired by the infamous 1960s LSD experiment with Tusko the Elephant. It’s a deranged, tragic, and true story, and I felt it deserved to be enshrined, or rather lampooned, in music.
I wrote most of the lyrics. I was learning the guitar. But I didn’t know about arranging, couldn’t sing very well, and didn’t have access to a studio. So ultimately, the notes got buried, and I moved on to other projects, like my PhD.
Then, recently, I dug them up again. I finished the lyrics and the chords. The strength of this piece was in the rhyme and story. The musical part was just to make it palatable. In my opinion, I’m a better lyricist than ChatGPT, so the text was man-made. A thesaurus and rhyming dictionary (all pre-2000 tech) were my stack.
Why revisit it? Because I was better at singing and accompanying myself on guitar by now? No. The reason was obvious: AI happened.
To take ideas from paper to audible, I’ve been using Suno. I recorded myself in a podcast studio, singer-songwriter style. Uploaded that as context, and iterated for days until I had the right vibe. (I was going for something like Carolina Drama by the Raconteurs, a side project of Jack White.) At first, the model wouldn’t sing the story with the right intonation and pauses. Those were crucial for the comedy to land. So I separated the backing track, sang over that for another 2 iterations, and did a lot of local edits and incisions. After more than a week, it was just right. I also put together a perfunctory video clip: another couple evenings. I used AI image generation for that too, editing video using Claude and Openshot.
I would never have been able to produce something of this level before. I had some, but still far from all, of the skills required to produce music video. That was no longer a show-stopper.
Proudly, I went to share my creation with the internet. At this point, I had put a lot of effort into this. Still only, say, 10% of the effort it used to take, but at least two weeks full-time.
The reactions were abysmal. No one cared for my clever lyrics. On the surface, it looked and sounded like slop. So, to the internet at large, it was slop. I only posted in two subreddits, but found myself banned from five. When I tried to plead my case, emphatically defending it was handwritten, but conceding AI assistance, I was banished from the site altogether. 2 / 5 stars, do not recommend.
Honestly, the song is quite cute. And it was demonstrably better as an AI product than it would have been with me singing into my phone like a scrub. Ironically, I suspect the latter would have been more successful.
Now, I’ll admit I am salty over being rejected, even by a notoriously vitriolic band of anonymous keyboard warriors. Moreover, Reddit specifically has a major problem with AI bots overrunning their platform and drowning out human creators. They are not taking half measures. So perhaps I just didn’t read the room.
But it got me thinking about a bigger question: Why do people HATE AI content SO much?
John Henry
There is a 19th century US folklore tale about a heroic freedman named John Henry. This steel-drivin’ man went head to head with a steam-powered rock drill, digging a railroad tunnel. He valiantly beat the machine, and promptly died. This heralded the age of industry.
If that sounds familiar, it’s because we’ve been here many times. I understand that people are worried. AI is making creatives and researchers, lawyers and copywriters obsolete, increasing inequality. It’s a tough pill to swallow, but I think the writing is on the wall. Change is coming, and we have to adapt. The cries of luddites won’t help: at best this is another moat that technology is backfilling, at worst these cries are gatekeeping. Good to note, I’ve been developing code for a living for the past 4 years.
Other people voice real concerns I share: AI’s power and water consumption. Or the existential threat that ASI poses to the order of humanity as we know it. These are good reasons to be skeptical of the wanton development of new models and the sprawl of datacenters.
But all of these seem misplaced, to me, as a critique of the content itself. That is either high or low quality. Useful or not. I have always believed more in the value of the message over the messenger. It’s why I gravitated to objective and falsifiable subjects like CS, math and physics. I understand not everyone feels that way. Still, I think there’s more to this: The soul.
Artists and Effort
I’ve been puzzled by a version of this question for a long time, long before ChatGPT. I actually started making digital art 8 years ago. Completely ‘white-box’, as they say, generative according to algorithms I wrote, and therefore instant. Though people agreed that the art was pretty, and liked that I could make a unique print for them on demand, most declined to pay for it. Because, ostensibly, no human effort had gone into making that individual piece.
I also paint. I have the dexterity of a child and use simple techniques, but worked for months on some pieces. A friend, the kind who speaks before she thinks (I’m actually grateful for such candor), saw one. Her house is full of art made by radical avant-garde nudists and shamanic sculptures. She told me, “It’s impressive. And if you hadn’t made it, I might buy it.”
It wasn’t for sale. And also, ouch! But then I embraced the truth. People want to tell a story to their dinner guests when they’re sipping wine and studying their home gallery. It has to mean something. People buy art, not firstly for its quality, but for the message of the artist. My message is kind of boring. I’m just a tech nerd who moonlights in neat patterns. I’m not a martyr and I don’t live my art.
I distilled two very important insights:
Before you disagree, just think about it. Industrial automation has slashed the prices of household products, while boosting quality. A printed photograph is a much higher-fidelity representation of any subject than an oil painting. But exactly those imperfections, the irregular strokes and dollops, are what tell us “someone slaved over this”. It’s priceless.
There’s a striking experiment by Banksy, who offered his ‘spray art’ incognito on the streets of NYC, for 60 USD per canvas. He sold 7. Later people kicked themselves, if they had known someone so famous had made it with his own hands, they would have realized it was worth a thousand times that. Why couldn’t they detect the value based on the art itself?
The breakdown cuts two ways. If I try to sell you a rock, you might pass, “it’s just a rock”. I counter, “But THIS rock was flipped over 10,000 times by pagan monks, once every full moon, for the last 808 years.” Do the scratches make the rock more valuable? Isn’t it still just a rock?
I hope this is not just my personal soapbox, trying to vindicate myself from my failed artist endeavors. Maybe my art sucks. But my point stands. Let me tie this into the larger story about GenAI.
You down with O.P.P? (Other People’s Prompts)
First off, I think my question needs a little clarification. Because we clearly don’t hate all AI content. We prompt and consume AI content all the time. We just really hate being served other people’s AI content. One internet pundit made the analogy that we also don’t mind our own humming or the smell of our own farts. There's actually a beautiful metaphor here. When we prompt the content, we know exactly where it comes from, and what it’s made of.
We don’t want to be made to guess where the stink comes from.
Before, if you wanted to publish an academic paper, you had to have:
Now, you can submit a paper while missing any number of these. Even if you don’t have a single one: “Chat, give me 10 novel ideas for math breakthroughs.”
So instead of being throttled to the population of authors that happened to be blessed with all 5, any subset will do. Before, polished research came with two implicit signals:
Now, things that look polished may not carry that seal of approval at all. And the number of papers submitted to journals has skyrocketed. Much is slop. But some is not. While it’s little consolation to the poor editors, I think the total amount of good research has gone up, while the proportion has gone down. Because, of course, not all good ideas happen to be born inside famous or prolific writers. Good ideas may not discriminate. Low effort does not prove low quality.
The same mechanism applied to my song. (I hope.)
The Burden Shifts
So what’s the problem? There is a huge influx of bad content. Also good content. Nowadays, both are polished. It’s getting harder and harder to tell what all that polish actually means. It’s the age of photography, and the value of lifelike portraits is plummeting hard.
As creation is becoming cheaper and cheaper, a larger proportion of our job is curation. Quality control. Trust in our sources is becoming more important. We need to cut out more and more noise, and AI content is giving us more of that job to do.
We had good methods for discerning real paint from photographs, real wood from laminate, real plants from plastic. We desperately want a way to tell GenAI from human-authored content. Because we don’t want to be the dunce who spends time reading something that the purported author didn’t even take the time to write. We want a guarantee: “trust me, this is worth your time”. We can’t trust anymore. The burden has shifted from creator to consumer. And we hate it.
I think we will gravitate towards a new culture. Anthropic is watermarking its text. It’s tricky for a human eye but easy to algorithmically detect GenAI video and imagery. Part of this is a legislative issue. And part is being solved organically. I actually kind of dig the trend of companies and figureheads in the AI community having barebones white HTML websites with no CSS. They are counter-signalling: “I didn’t use Claude for this. It’s artisanal”.
But I also think we can be more open-minded. ChatGPT is solving math problems. Those theorems are no less true. I personally listen to some GenAI music. It’s good. A lot of value is being created. Yes, it was produced by mindless automatons who cannot morally or in good conscience vouch for its meaning or profundity. But those automatons were created by humans with vision and unimaginable resources.
Going forward, most good content will be a collaboration between humans and agents. It’s a massive enabler and productivity booster. We need to institute new systems for filtering. Sources that consistently serve their audience worthwhile ideas, regardless of the mode of creation, should be rewarded with trust. So then, let’s take a moment to appreciate the wealth of new tools, art, and science that can benefit us all.
Coda
Actually, I told one lie at the beginning.
This essay was one-shot by Claude.
How do you feel about the opinions espoused within? Are they less valid?
.
.
.
Nah, just kidding. I've learned my lesson, it's all Daniel.