AI generated web content has got to be one of the most counterproductive things to use AI on.
If I wanted an AI summary of a topic or answer to a question, a chatbot of choice can easily provide that for you. There’s no need for yet another piece of blogspam that isn’t introducing new information into the world. That content is already available inside the AI model. At some point, we’ll get so oversaturated with fake, generated BS that there won’t be enough high quality new information to feed them.
This is the fundamental reason why I am in favor of a ban on simply posting AI-generated content in user forums. It isn't that AI is fundamentally bad per se, and to the extent that it is problematic now, that badness may well be a temporary situation. It's because there's not a lot of utility in you as a human being basically just being an intermediary to what some AI says today. Anyone who wants that can go get it themselves, in an interactive session where they can explore the answer themselves, with the most up-to-date models. It's fundamentally no different than pasting in the top 10 Google results for a search with no further commentary; if you're going to go that route just give a letmegooglethat.com link. It's exactly as helpful, and in its own way kind of carries the same sort of snarkiness with it... "oh, are you too stupid to AI? let me help you with that".
Similarly, I remember there was a lot of frothy startup ideas around using AI to do very similar things. The canonical one I remember is "using AI to generate commit messages". But I don't want your AI commit messages... again, not because AI is just Platonically bad or something, but because if I want an AI summary of your commit, I'd rather do it in two years when I actually need the summary, and then use a 2027 AI to do it rather than a 2025 AI. There's little to no utility to basically caching an AI response and freezing it for me. I don't need help with that.
It's been interesting to watch this play out in microcosm in different spaces. Danbooru and Gelbooru are two anime image boards that banned AI image content, largely to their benefit in my opinion. Rule34 is a similar image board that has allowed AI images and they've need to make tagging and searching adaptations to add to handle the high volume of AI images versus human artists. I'm glad there's an ecosystem of different options, but I find myself gravitating to the ones that have banned AI content.
I fully agree with this, besides that if an AI could auto-generate a commit message that I can edit to make actually correct and comprehensive, it will probably be a better, more descriptive message than whatever I come up with in usually ~3 minutes.
The value is a nice starting point but the message is still confirmed by the actual expert. If it's fully auto-generated or I start "accepting" everything, then I agree it becomes completely useless.
> It's because there's not a lot of utility in you as a human being basically just being an intermediary to what some AI says today.
To be fair, there has never been a lot of utility in you as a human being involved, theoretically speaking. The users do not use a forum because you, a human, are pulling knobs and turning levers somewhere behind a meaningless digital profile. Any human involvement that has been required for the software to function is merely an implementation detail. The harsh reality, as software developers continually need to be reminded of, is that users really don't care about how the software works under the hood!
For today, a human posting AI-generated content to a forum is still providing all the other necessary functions required, like curation and moderation. That is just as important and the content itself, but something AI is still not very good at. A low-value poster may not put much care into that, granted, but "slop" would be dealt with the same way regardless of whether it was generated by AI or hand written by a person. The source of content is ultimately immaterial.
Once AI gets good, we'll all jump to AI-driven forums anyway, so those who embrace it now will be more likely to stave off the Digg/Slashdot future.
I notice this idea repeatedly popping up, that "content made by humans" is more-or-less equivalent to content generated by genAI inasmuch as genAI can "in theory" (in some sense) create the same thing a human could and so they're both "in theory" the same thing.
The idea that words people write don't mean anything or imply anything in an abstract sense is misguided, in my opinion. When one reads something a person wrote, they think about what the person who wrote it was thinking, what it means to them, what the implications of what they think might be... there are people who do not think about things like this, so they don't care and view genAI text as equivalent because that level of thought simply isn't put into their reading.
Anyway, my point is talking on a forum filled with LLMs would probably stop being interesting and engaging very quickly because LLMs are bad at emulating the lateral thinking, diversity of ideas, and abstraction of communication that make talking to a human fun.
> because LLMs are bad at emulating the lateral thinking, diversity of ideas, and abstraction of communication that make talking to a human fun.
While that is no doubt true today, the earlier comment that sparked this posits that may only be a temporary state that may improve in the future. Once LLM and human creation is indistinguishable, there is no reason to have concern for what generated the content, is there?
Nobody uses a forum for the human connection. There is no human! I can't see your face, I can't touch your skin, I don't feel the heat radiating from your body. Hell, if we meet each other on the street later today, I'll never know it was you. There is only software. I do assume, knowing a thing or two about how the technology works, that in implementation that there is a human somewhere in the loop, but I don't completely know for sure, and it wouldn't make a difference anyway.
There is a place for human connection, most certainly, but it is found in the "real world". Forums are not equivalent. They are something else entirely.
> Anyway, my point is talking on a forum filled with LLMs would probably stop being interesting and engaging very quickly
Just as it does when humans write drivel by hand. There is merit to banning accounts that post garage, but what produced that garbage is irrelevant. AI's involvement, or lack thereof, makes no difference. The quality of an account can be judged on its output, not the mechanism by which it operates.
> Just as it does when humans write drivel by hand.
I see this a lot in AI discussions - it can, at best, do what we do at a level we'd consider "good enough." It can write mediocre slop just as well as the most mediocre of us. To me, that is lowering the bar for our work exceptionally. We shoot for the stars; we just miss sometimes!
You don't have to lower the bar in order to accept AI, but where AI is able to meet the bar, there is no logical reason to ban it. The user really doesn't care about how the software is implemented.
I understand the luddites are always fearful of losing their job, but that is ultimately irrational.
I don't think better search is exactly what we want. It would also be great to have less quantity and more quality. I think optimizing only search to make it better (including AI) only furthers the quantity aspect of content, not quality. Optimizing search or trying to make it better is the wrong goal IMO.
Arguably, the opposite is true. Ars Technica and others have written about this extensively [0].
Having summarized results appear immediately with links to the sources is preferable to opening multiple tabs and sifting through low-quality content and clickbait.
Many real-world problems aren't as simple as "type some keywords" and get relevant results. AI excels as a "rubber duck", i.e., a tool to explore possible solutions, troubleshoot issues, discover new approaches, etc.
Yes, LLMs are useful for junior developers. But for experienced developers, they're more valuable.
It's a tool, just like search engines.
Airplanes are also a tool. Would you limit your travel to destinations within walking distance? Or avoid checking the weather because forecasts use Bayesian probability (and some mix of machine learning)? Or avoid power tools because they deny the freedom of doing things the hard way?
One can imagine that when early humans began wearing clothing to keep warm, there were naysayers who preferred to stay cold.
The most creative people I know are using AI to further their creativity. Example: storytelling, world building, voice models, game development, artwork, assistants that mimic their personality, helping loved ones enjoy a better quality of life as they age, smart home automations to help their grandmother, text-to-speech for the visually impaired or those who have trouble reading, custom voice commands, and so on.
Should I tell my mom to turn off Siri and avoid highlighting text and tapping "Speak" because it uses AI under the hood? I think not.
They embrace it, just like creative people have always done.
Socrates had a skeptical view of written language, preferring oral communication and philosophical inquiry. This perspective is primarily presented through the writings of his student, Plato, particularly in the dialogue Phaedrus.
I confirmed that from my own memory via a Google AI summary, quoted verbatim above. Of course, I would never have learned it in the first place had somebody not written it down.
> Socrates had a skeptical view of written language, preferring oral communication and philosophical inquiry. This perspective is primarily presented through the writings of his student, Plato, particularly in the dialogue Phaedrus.
He did not. You should read the dialogue.
> I confirmed that from my own memory via a Google AI summary, quoted verbatim above.
This is the biggest problem with LLMs in my view. They are great at confirmation bias.
In Phaedrus 257c–279c Plato portrays Socrates discussing rhetoric and the merits of writing speeches not writing in general.
"Socrates:
Then that is clear to all, that writing speeches is not in itself a disgrace.
Phaedrus:
How can it be?
Socrates:
But the disgrace, I fancy, consists in speaking or writing not well, but disgracefully and badly.
Phaedrus:
Evidently."
I mean, writing had existed for 3 millennia by the point this dialogue was written.
This only means that the web (websites and web 2.0 platforms) for public usage is becoming redundant because any type of data that can be posted on the web can now be generated by an LLM. LLMs have been only around for a short while but the web is already becoming infested with AI spam. Future generations that are not accustomed to the old pre AI web will prefare to use AI rather than the web, LLMs will eventually be able to generate all aspects of the web. The web will remain useful for private communication and general data transfer but not for surfing as we know it today.
Edit to add:
Projects like the Internet Archive will be even more important in the future.
Editorial guidelines at many publications explicitly state that AI can assist with drafts, outlines, and editing, but not with generating final published stories.
AI is widely used for support tasks such as:
- Transcribing interviews
- Research assistance and generating story outlines
- Suggesting headlines, SEO optimization, and copyediting
- Automating routine content like financial reports and sports recaps
This seems like a reasonable approach, but even so I agree with your prediction that people will mostly interact with the web via their AI interface.
Anthropic cares about that, every individual content creator does not. Their goal is to win the war for attention, which is now close to zero sum with everyone on the internet and there's only 24 hours in the day.
> AI generated web content has got to be one of the most counterproductive things to use AI on.
For something like a blog I would agree, but I found AI to be fantastic at generating copy for some SaaS websites I run. I find it to be a great "polishing engine" for copy that I write. I will often write some very sloppy copy that just gets the point across and then feed that to a model to get a more polished version that is geared to a specific outcome. Usually I will generate a couple variants of the copy I fed it, validate it for accuracy, slap it into my CMS and run an a/b test and then stick with the content that accomplishes the specific goal of the content best based on user engagement/click through/etc.
While I largely agree, I don't think it's quite correct to say AI generated blogs contain no new information. At least not in a practical sense. The output is a function of the the LLM and the prompt. The output contains new information assuming the prompt does. If the prompt/context contains internal information no one outside the company has access to, then a public post thus generated certainly contains information new to the public.
Using AI generated content to mass-scale torpedo the web could be a tool to get people off of Google and existing social media platforms.
I'm certainly using Google less and less these days, and even niche subreddits are getting an influx of LLM drivel.
There are fantastic uses of AI, but there's an over-abundance of low-effort growth hacking at scale that is saturating existing conduits of signal. I have to wonder if some of this might be done intentionally to poison the well.
> Using AI generated content to mass-scale torpedo the web could be a tool to get people off of Google and existing social media platforms.
How? Fill the web with AI generated content or just using LLMs to search for information? As more junk is poured into training LLMs this too will take a hit at some point. I remember how great the early web search was, one could find from thousands to millions of hits for request. At some point it got so polluted that it became nearly useless. It wasn't only spam that made is less useful, it was also the search providers who twisted the rules to get them to reap all the benefits.
This is pretty reductive. Many people want to pump some new thoughts they had into an AI to generate something tolerable to post on their blog. The writing isn't the point; the thoughts are. But they can't just post 200 words of bullet points (or don't feel like they can, anyway). So the AI is an assistant which takes their thoughts and makes them look acceptable for publication.
I find it interesting that the 6 replies to my comment assumed that I was talking about myself, when in fact my comment specifically was talking about other people. I can and do write just fine, thank you. But I have sympathy for people who have something to say and struggle to say it, whether because they lack the skill or the time or the executive function. They don't need to be recognized as a writer, they just want to have an impact.
Using LLMs to write is like wearing fast fashion. If you want certain kinds of people to notice you and think of you favorably, you need to wear clothes that represent how you want to be seen. It would be better if people made their own clothes, or paid money to "real" designers for bespoke and/or high-quality items. But most people can't afford the time and money for that. So we have places like Hot Topic and Forever 21 and you probably recognize when other people shop there and you turn up your nose at them. But they are still effective at getting what they want from their sartorial choices.
Your second paragraph makes no sense, other than to make the expectation of effort sound like classism. Wearing clothes other people made is not a summary of your ideas, it's an expression of your aesthetic.
Writing is free. About as free as anything can be. It's not classism to ask people to put some actual thought into their ideas. It's just reality that people won't want to read slop.
>I find it interesting that the 6 replies to my comment assumed that I was talking about myself
No, it's just colloquial English on a forum to write you as an address to the reader, not literally to the person you reply to. As in, "if you (someone) feels they can't write without AI, then writing may not be for you (that person)."
if you can’t write your thoughts as something cohesive to begin with i don’t using LLMs is going to solve your problem. writing is absolutely the point if you’re trying to communicate with text. lack of clarity is usually sign of lack of understanding imo, i see it in my own writing
This is so, so wrong. The writing is the thoughts. A person's un-articulated bullet points are not worth that much. And AI is not going to pull novel ideas out of your brain via your bullet points. It's either going to summarize them incorrectly or homogenize them into something generic. It would be like dropping acid with a friend and asking ChatGPT to summarize our movie ideas.
The idea that writing is an irrelevant way to gatekeep people with otherwise brilliant ideas is not reality. You don't have to be James Baldwin, but I will not get a sense for what your ideas even are via an AI summary.
> The writing isn't the point; the thoughts are. But they can't just post 200 words of bullet points (or don't feel like they can, anyway).
Who or what is clamoring for that AI-generated padding which turns 200 words of bullet points into 2000 words of prose, though? It's not like there's suddenly going to be 10x more insight, it's just 10x more slop to slog through that dilutes whatever points the writer had.
If you have 200 words' worth of thoughts you want to share... you can just write 200 words.
The writing is the point. A well-structured, well-argued, and well-written article indicates the writer has devoted considerable time to understanding and thinking through the topic — if they haven't, it quickly becomes obvious. A series of bullet points indicates the opposite, and using an AI to hide the fact that the "writer" has invested minimal cognitive effort is dishonest.
It's fascinating how creative these large AI companies are at finding ways to burn through VC funding. Hire a team of developers/content writers/editors, tune your models, set up a blog and build an entire infrastructure to publish articles to it, market it, and then...shut it all down in a week. And this is a company burning through multiple billions of dollars every quarter just to keep the lights on.
I've always thought of these VC fueled expeditions to nowhere as the opposite. Wealth transfer from the owning class to the middle class seeing as a lot of these ventures crash and burn with nothing to show for it.
Except for the founders/early employees who get a modest (sometimes excessive) paycheck.
> I've always thought of these VC fueled expeditions to nowhere as the opposite. Wealth transfer from the owning class to the middle class seeing as a lot of these ventures crash and burn with nothing to show for it.
That would be the case if VCs were investing their own money, but they're not. They're investing on behalf of their LPs. Who LPs are is generally an extremely closely-guarded secret, but it includes institutional investors, which means middle-class pensions and 401(k)s are wrapped up in these investments as well, just as they were tied up in the 2008 financial crisis.
I think the chilling effect on mom and pop businesses undoes all of that. When they (we) disrupt and industry the power consolidates but in new hands. The idea is to get it away from the entrenched interests but like a good cultural revolution the second tier ends up in charge when the first tier gets beheaded.
What makes you think they think that? If someone says “finding creative ways to murder people” you think they’re saying the problem is the “creative” part?
The problem with using AI for writing is that most of the time you're trying to convey some kind of information that the AI doesn't know. If your business has outperformed some metric, the AI doesn't know until you tell it. So unless you write a very long prompt with all the facts and data that you want to convey, you just get fluff. If you do that, you get pretty polished prose but it doesn't save you all that much time.
Is there an archive anywhere? People can argue to no end based on some whimsical assumptions of what the blog was and why it was taken down, but it really comes down to the content. I have found even o3 cannot write high-quality articles on the topics I want to write about.
Have you tried Perplexity's Discover feed? It's my go-to source of news these days. I don't know what model they use to generate content but it's really good.
Did the reporter reach out to Anthropic for public comment on this? They list a "source familiar" with some details about what the intended purpose was for, but no mention on the why
Up until a few weeks ago, my LinkedIn seemed to become better because of AI, but now it seems everything is lazy AI slop.
We meatbags are great pattern recognizers.
Here is a list of my current triggers:
"The twist?",
"Then something remarkable happened",
That said, this is more of an indictment of the lazyness of the authors to provide clearer instructions on the style needed so the app defaults to such patterns.
These are the first four words of each sentence from a thing ChatGPT wrote. I will let this speak for itself, mostly. Can you see the forest past the trees and spot the major "turning points"?
(Context: I told it to write at "will" after a session of explaining Rene Girard's mimetic desire in the styles of various authors)
*Well, here we are,
You've got me tangled
*And hey, maybe I
*Imagine this: I'm sitting
*And maybe now, I'm
You've got me thinking
*Maybe it's saying, "Hey,
*And so, I think
Not just any wanting,
The kind that makes
There's something beautiful about
The way it drives
*But here's the kicker:
It's got a mind
It makes us do
It's a double-edged sword
*And yet, without it,
Probably just sitting around,
*So maybe, just maybe,
*Not just because you've
*It's that tiny ember
In the end, desire's
It's the fire that
*And maybe, just maybe, [yes, again]
So here's to the
May it burn bright
>This is one of the usual key turning points in these essays. An earlier one happened when it was like [introduces idea] [straw-mans objection] [denies strawman]. I didn't bring it up because it's not always a strong one and this one didn't seem entirely too heavy-handed. There is, however, much more often a very obvious "But here's the thing" (or similar) to be found. As soon as I saw that, I already knew I was going to find a paragraph beginning with "So" somewhere near the end.
I can tell you this much - most people who are opposed to AI writing blog articles are usually from the editorial team. They somehow believe they're immune to being replaced by AI. And this stems from the misconception that AI content will always sound AI, soul-less, dry, boring, easy to spot and all that. This was true with ChatGPT-3xx. It's not anymore. In fact, the models have advanced so much so that you will have a really hard time distinguishing between a real writer and an AI. We actually tried this with a large Hollywood publisher in-house as a thought experiment. We asked some of the naysayers from the editorial + CXO team to sit in a room with us, while we presented on a large white screen - a comparison of two random articles - one written by AI, which btw wasn't trained, but just fed a couple of articles of the said writer on the slide into the AI's context window, and another which was actually written by the writer themselves. Nobody in the room could tell which was AI and which wasn't. This is where we stand today. Many websites you read daily actually have so much AI in them, just that you can't tell anymore.
Counterpoint: GPT-4 and later variants, such as o3 and 4.5, have such a characteristic style that it's hard not to spot them.
Em dashes, "it's not just (x), it's (y)," "underscoring (z)," the limited number of ways it structures sentences and paragraphs and likes to end things with an emphasized conclusion, and I could go on all day.
DeepSeek is a little bit better at writing in a generic and uncharacteristic tone, but still... it's not good.
If you ask them to speak in a different voice, they will. It's only characteristic if the user has made no effort at all to mask that it is AI generated content.
> We asked some of the naysayers from the editorial + CXO team to sit in a room with us, while we presented on a large white screen - a comparison of two random articles - one written by AI, which btw wasn't trained
Needlessly close to bullying way to try and prove your point.
Well, at least it passes the basic sanity test I've been using. ZeroGPT, on the other hand, gives a "100% AI" rating to the introduction and preamble of the US Declaration of Independence (from "When in the Course of human events" to "provide new Guards for their future security").
Yep, it is not able to recognize. To be fair, it's not just dump it into ChatGPT and copy paste kind of AI. We feed it into the model in stages, we use 2-3 different models for content generation, and another 2 later on to smoothen the tone. But, all of these are just OTS models, not trained. For example, we do use Gemini in one of the flows.
If I wanted an AI summary of a topic or answer to a question, a chatbot of choice can easily provide that for you. There’s no need for yet another piece of blogspam that isn’t introducing new information into the world. That content is already available inside the AI model. At some point, we’ll get so oversaturated with fake, generated BS that there won’t be enough high quality new information to feed them.
reply