PhysicsForums and the Dead Internet Theory
hallofdreams.org
hallofdreams.org
(I mean "nobody" in the sense of "nobody likes Nickelback". ie, not literally nobody.)
If I want to talk to an AI, I can talk to an AI. If I'm reading a blog or a discussion forum, it's because I want to see writing by humans. I don't want to read a wall of copy+pasted LLM slop posted under a human's name.
I now spend dismaying amounts of time and energy avoiding LLM content on the web. When I read an article, I study the writing style, and if I detect ChatGPTese ("As we dive into the ever-evolving realm of...") I hit the back button. When I search for images, I use a wall of negative filters (-AI, -Midjourney, -StableDiffusion etc) to remove slop (which would otherwise be >50% of my results for some searches). Sometimes I filter searches to before 2022.
If Google added a global "remove generative content" filter that worked, I would click it and then never unclick it.
I don't think I'm alone. There has been research suggesting that users immediately dislike content they perceive as AI-created, regardless of its quality. This creates an incentive for publishers to "humanwash" AI-written content—to construct a fiction where a human is writing the LLM slop you're reading.
Falsifying timestamps and hijacking old accounts to do this is definitely something I haven't seen before.
So far (thankfully) I've noticed this stuff get voted down on social media but it is blowing my mind people think pasting in a ChatGPT response is productive.
I've seen people on reddit say stuff like "I don't know but here's what ChatGPT said." Or worse, presenting ChatGPT copy-paste as their own. Its funny because you can tell, the text reads like an HR person wrote it.
The ones that make me furious are on some of the mental health subreddits. People are asking for genuine support from other people, but are getting AI slop instead. If someone needs support from an AI (which I've found can actually help), they can go use it themselves.
You should have a lot less confidence in your ability to discern what's AI generated content, honestly. Especially in such contexts where the humans will likely be writing very non-offensive in order to not-trigger the OP.
It makes me wonder how shallow a person's knowledge of all areas must be that they could use an LLM for more than a little while without encountering something where it is flagrantly wrong yet continued with its same tone of absolute confidence and authority. ... but it's mostly just a particularly aggressive form of Gell-Mann amnesia.
Hopefully to everyone on HN, but definitely not to everyone on the greater Internet. There are plenty of horror stories of people who apparently 100% blindly trust whatever ChatGPT says.
Then 10 hours later OP edited the post and dropped the bomb. The screenshot of their prompt “make a story for the am I the asshole subreddit that makes a fat person look bad.” Followed by the post they pasted directly from chatgpt. Only one comment was about the edit and it completely missed the point and instead blamed OP for tricking them. Not the fact that probably every post on that subreddit and others like it is AI slop.
It makes perfect intuitive sense if you don't know how the things actually work.
It's not just generated content. This problem has been around for years. For example, google a recipe. I don't think the incentives are there yet. At least not until Google search is so unusable that no one is buying their ads anymore. I suspect any business model rooted in advertising is doomed to the eventual enshitification of the product.
This reminds me of the time around ChatGPT 3's release where Hacker News's comments was filled with users saying "Here's what ChatGPT has to say about this"
(I very much would like any AI generated text to be marked as such, so I can set my trust accordingly)
Depends on what they are used for and what they are purporting to represent.
For example, I really hate AI images being put into kids books, especially when they are trying to be psuedo-educational. A big problem those images have is from one prompt to the next, it's basically impossible to get consistent designs which means any sort of narrative story will end up with pages of characters that don't look the same.
Then there's the problem that some people are trying to sell and pump this shit like crazy into amazon. Which creates a lot of trash books that squeeze out legitimate lesser known authors and illustrators in favor of this pure garbage.
Quite similar to how you can't really buy general products from amazon because drop shipping has flooded the market with 10 billion items with different brands that are ultimately the same wish garbage.
The images can look interesting sometimes, but often on second glance there's just something "off" about the image. Fingers are currently the best sign that things have gone off the rails.
Sometimes this comment gets a ton of upvotes. Sometimes it gets indignant replies insisting it's real writing. I need to come up with a good standard response to the latter.
How about, "I'm sorry, but if you're willing to use AI image slop, how should I know you wouldn't also use AI text slop? AI text content isn't reliable, and I don't have time to personally vet every assertion."
Also, "enemy"? That's a little harsh, don't you think? I would never consider a random doofus on an internet forum to be my enemy.
It sucks, but it doesn't suck any more than what was done in the past: Litter the article with stock photos.
Either have a relevant photo (and no, a post about cooking showing an image of a random kitchen, set of dishes, or prepared food does not count), or don't have any.
The only reason blog posts/articles had barely relevant stock images was to get people's attention. Is it any worse now that they're using AI generated images?
wait for the proof-of-humanity decade where you're paid to be here and slow and flawed
Once you have people sorting through them, editing them, and so on the curation adds enough additional interest...and for many people what they get out of looking at a gallery of AI images is ideas for what prompts they want to try.
The other half of the problem is that rephrasing information doesn't actually introduce new information. If I'm looking for the kind of oil to use in my car or the recipe for blueberry muffins, I'm looking for something backed by actual data, to verify that the manufacturer said to use a particular grade of oil or for a recipe that someone has actually baked to verify that the results are as promised. I'm looking for more information than I can get from just reading the sources myself.
Regurgitating text from other data sources mostly doesn't add anything to my life.
If llms could take the giant overwhelming manual in my car and get out the answer to what oil to use, that woukd be useful and not new information
You can literally just google that or use the appendix that's probably at the back of the manual. It's also probably stamped on the engine oil cap. It also probably doesn't matter and you can just use 10w40.
Reminds me of the old Yogi Berra quote: Nobody goes there anymore, its too crowded.
I basically decided that using AI content would waste everyone's time. However, it's a real chicken-or-egg problem in content creation. Faking it to the point of project viability has been a real issue in the past (I remember the reddit founders talking about posting fake comments and posts from fake users to make it look like more people were using the product). AI is very tempting for something like this, especially when a lot of people just don't care.
So far I've stuck to my guns, and think that the key to a course wiki is absolutely having locals insight into these courses, because the nuance is massive. At the same time, I'm trying to find ways that I can reduced the friction for contributions, and AI may end up being one way to do that.
Of the top of my head I wonder if there's a way to have AI generate a summary from existing (on-line) information about a course with a very explicit "this is what AI says about this course" or some similar disclosure until you get 'real' local insight. No one could then say 'it's just AI slop', but you're still providing value as there's something about each course. As much as I personally have reservations about AI, I (personally, YMMV) am much more forgiving if you are explicit about what's AI and what's not and not trying to BS me.
I've always insisted that if it is financially feasible, I'd want the app to become a 501(c)(3) or at least a B-Corp, maybe even sold to Wikimedia. Still, the number of people who contribute to the side vs the number who visit is somewhere in the range of 1:10,000 (if that) right now, so concern about offending contributors is non-trivial.
As it stands, I've generally gone to the courses' sites and just quoted what they have to say about their own course, but that really isn't what I want to do, even if it is generally informative. Unfortunately, there is rarely hole-by-hole information, which is the level of granularity I'm going for.
The general trend of viewing LLM features as forced against users' will and the now widespread use of "slop" as a derogatory description seems to indicate the general public is less enthusiastic about these consumer advances than, say, programmers on HN.
I use LLMs for programming (and a few other, general QA things before a search engine/wikipedia visit) but want them absolutely nowhere else (except CoPilot et al in certain editors)
This timeline tracks with my own blogging. Google slowly stopped ranking traditional forum posts and blogs as well around that time, regardless of quality, unless it was a “major”.
> But, unlike so many other fora from back in the early days, it went from 2003 to 2025 without ever changing its URLs, erasing its old posts, or going down altogether.
I can also confirm if you have a bookmark to my blog from 2008, that link will still work!
The CMS is no longer, it's all static now... which too few orgs take the short amount of time to bother with when "refreshing" their web presence :(
IMO the true inflection point was 2014 when Google first hid (from the UI) and then fully removed (no longer accessible by magic URL) the “Blogs” and especially the “Discussions” filters. Some contemporary discussions on “Discussions”:
- https://techcrunch.com/2014/01/23/googles-search-filters-now...
- http://googlesystem.blogspot.com/2014/03/bring-back-forum-se... (details the briefly-working magic URLs)
- https://www.ghacks.net/2014/01/23/search-discussions-blogs-p...
- https://www.seroundtable.com/google-search-filters-gone-1799...
- https://www.webmasterworld.com/google/4687960.htm
- https://www.thecoli.com/threads/i-cant-google-search-by-disc...
- https://www.neogaf.com/threads/anyone-else-annoyed-google-re...
- https://webapps.stackexchange.com/questions/57249/has-the-op...
- https://www.bladeforums.com/threads/how-to-do-google-discuss...
- https://browsermedia.agency/blog/alternatives-discussion-sea...
The option appeared randomly for me on a search, and I took immediately note of the udm number. :)
8 = jobs (but doesn't return any results) 15 = attractions (but doesn't return any results)
> The backdated answers were an internal test. We conceived of a bot that would provide a quality answer to a thread without a reply after 1+ years. That too also failed. Instead, I’m considering pruning all threads without a reply as they clutter up the forums.
> There’s also a social contract: when we create an account in an online community, we do it with the expectation that people we are going to interact with are primarily people. Oh, there will be shills, and bots, and advertisers, but the agreement between the users and the community provider is that they are going to try to defend us from that, and that in exchange we will provide our engagement and content. This is why the recent experiments from Meta with AI generated users are both ridiculous and sickening. When you might be interacting with something masquerading as a human, providing at best, tepid garbage, the value of human interaction via the internet is lost.
It is a disaster. I have no idea how to solve this issue, I can't see a future where artificially generated slop doesn't eventually overwhelm every part of the internet and make it unusable. The UGC era of the internet is probably over.
For readers who are not among the cognoscenti on the topic: in 1997 supercomputers started playing chess at around the same level as top grandmasters, and some PCs were also able to be competitive (most notably, Fritz beat Deep Blue in 1995 before the Kasparov games, and Fritz was not a supercomputer). From around 2005, if you were interested in chess, you could have an engine on your computer that was more powerful than either you or your opponent. Since about 2010, there's been a decent online scene of people playing chess.
So the chess world is kinda what the GPT world will be, in maybe 30ish years? (It's hard to compare two different technology growths, but this assumes that they've both hit the end of their "exponential increase" sections at around the same time and then have shifted to "incremental improvements" at around the same rate. This is also assuming that in 5-10 years we'll get to the "Deep Blue defeats Kasparov" thing where transformer-based machine learning will be actually better at answering questions than, say, some university professors.)
The first thing is, proving that someone is a person, in general, is small potatoes. Whatever you do to prove that someone is a real person, they might be farming some or all of their thought process out to GPT.
The community that cares about "interacting with real humans" will be more interested in continuous interactions rather than "post something and see what answers I get," because long latencies are the places where GPT will answer your question and GPT will give you a better answer anyways. So if you care about real humanity, that's gonna be realtime interaction. The chess version is, "it's much harder to cheat at Rapid or Blitz chess."
The second thing is, privacy and nonprivacy coexist. The people who are at the top of their information-spouting games, will deanonymize themselves. Magnus Carlsen just has a profile on chess.com, you can follow his games.
Detection of GPT will look roughly like this: you will be chatting with someone who putatively has a real name and a physics pedigree, and you ask them to answer physics questions, and they appear to have a really vast physics knowledge, but then when you ask them a simple question like "and because the force is larger the accelerations will tend to be larger, right?" they take an unusually long time to say "yep, F = m a, and all that." And that's how you know this person is pasting your questions to a GPT prompt and pasting the answers back at you. This is basically what grandmasters look for when calling out cheating in online chess; on the one hand there's "okay that's just a really risky way to play 4D chess when you have a solid advantage and can just build on it with more normal moves" -- but the chess engine sees 20 moves down the road beyond what any human sees, so it knows that these moves aren't actually risky -- and on the other hand there's "okay there's only one reason you could possibly have played the last Rook move, and it's if the follow up was to take the knight with the bishop, otherwise you're just losing. You foresaw all of this, right?" and yet the "person" is still thinking (because the actual human didn't understand why the computer was making that rook move, and now needs the computer to tell them that the knight has to be taken with the bishop as appropriate follow-up).
Honestly, (even) in my area of expertise, if the "abstraction/skill level" or the kind of wording (in your example: much less scientifically precise wording, "more like a 10 year old child asks"), it often takes me quite some time to adjust (it completely takes me out of my flow).
So, your criterion would yield an insane amount of false positives on me.
But Facebook already proved otherwise.
I am hoping OpenID4VCI[0] will fill this role. It looks to be flexible enough to preserve public privacy on forums while still verifying you are the holder of a credential issued to a person. The credential could be issued from an issuer that can verify you are an adult (banks) for example. Then a site or forum etc, that works with a verifier that can verify whatever combination of data of one or more credentials presented. I haven't dug into the full details of implementation and am skimming over a lot but that appears to be the gist of it.
[0] https://openid.net/specs/openid-4-verifiable-credential-issu...
She could start posting LLM nonsense, but people will be quick to point it out, and start unfollowing. An important part is that there’s no algorithm deciding what I see in my feed (unless I choose so), so random LLM stuff can’t really get into my feed, unless I chose so.
Another option is zero knowledge identity proofs that can be used to attest that you’re a human without exposing PII, or relying on a some centralized server being up to “sign you in on your behalf”
https://zksync.mirror.xyz/kWRhD81C7il4YWGrkDplfhIZcmViisRe3l...
EDIT: Sorry, I didn’t answer your question directly. So it doesn’t, but makes spam more expensive.
A lot or most of people will adapt, accept these conditions because compared to the constant threat of misery and precarity of work, or whatever other way to sustenance and housing, it will be very tolerable. Similar to how so called talk shows flourished, where fake personas pretend to get to know other fake personas they are already very well acquainted with and so on, while selling slop, anxieties or something. Like Oprah, the billionaire.
I have heard of Discord servers where admins won't assign you roles giving you access to all channels unless you've personally met them, someone in the group can vouch for you, or you have a video chat with them and "verify."
This is the future. We need something like Discord that also has a webpage-like mechanism built into it (a space for a whole collection of documents, not just posts) and is accessible via a browser.
Of course, depending on discovery mechanisms, this means this new "Internet" is no longer an easy escape from a given reality or place, and that was a major driver of its use in the 90's and 00's - curious people wanting to explore new things not available in their local communities. To be honest, the old, reliable Google was probably the major driver of that.
And it sucks for truly anti-social people who simply don't want to deal with other people for anything, but maybe those types will flourish with AI everywhere.
If the gated hubs of a possible new group-x-group human Internet maintain open lobbies, maybe the best of both worlds can be had.
I mention Discord because a lot of people use it for stuff that would formerly be forums. Telegram is also the the same. They're doing this despite it being centralized. What's the decentralized equivalent of Discord or Telegram? Does it support phone notifications?
https://news.ycombinator.com/item?id=35050858
(Even when something like a wiki exists, most of actual information will still be contained in the lore, itself blackholed by a deep web platform like Discord.)
Another option is a web of trust.
It's finally the year of gpg!
> We reached out to Greg Bernhardt asking for comment on LLM usage in PhysicsForums, and he replied:
> "We have many AI tests in the works to add value to the community. I sent out a 2024 feedback form to members a few weeks ago and many members don’t want AI features. We’ll either work with members to dramatically improve them or end up removing them. We experimented with AI answers from test accounts last year but they were not meeting quality standards. If you find any test accounts we missed, please let me know. My documentation was not the best."
Why they would recycle old human accounts as AI "test accounts", I have no idea.
https://www.linkedin.com/in/gregbernhardt
https://www.physicsforums.com/insights/author/greg-bernhardt...
https://x.com/GregBernhardt4/status/1875287174205374533
> "The dead internet theory is coming to fruition. This is a large reason I'm starting to cut back on social media and take back my time."
Presumably it's being done by the site-owner, whether that means new-management or original management getting desperate/greedy.
It all seems so unthinkable but when running a forum or a blog with an active comment section.. what would you do/think if your users show up, browse around and not say anything for a week? You start out by making topics in your own name, write helpful replies.. until you look like an idiot talking to yourself.
Forums with good traffic and lots of spammy advertisement no doubt consider it when visitors leave because nothing new happened.
I once upon a time, on a rather stale forum, created two similarly named accounts from the same ip and argued with myself. At first I thought the owner or one of the other users would notice but I quickly learned that no behaviour is weird enough for it to be ever considered.
Issues of trust and attribution have always existed for the web, but for many reasons it feels so much worse now--how bad must it get before some kind of sea-change can occur?
I'm not sure what the solution would be here.
* Does one need to establish a friggin' trademark for their own name/handle [0], just so they can threaten to sue using money they probably don't have?
* Is it finally time for PKI and everybody signs their posts with private keys and wastes CPU cycles verifying signatures to check for impersonation?
* Is there some set of implied collective expectations which need to be captured and formalized into the global patchworks of law?
[0] Ex: By establishing a small but plausible "business" selling advice and opinions under that name, and going after the impersonator for harming that brand.
If all else fails, there is always the web of trust (i think web of trust has a lot of issues, but establishing soneone is human seems like a much lower bar than establishing identity)
I have always wondered if people could attach some sort of cryptographic marker to their posts, that could link to an archive somewhere. Mostly I was thinking of backups of posts to yelp that couldn't be taken down, but I wonder if it would work that posts someone never made.
I expect the bad-actors will feed it into an LLM and say: "Rephrase this slightly", and they will get away with it because the big-money hucksters will have already convinced courts to declare it transformative or fair-use.
Or do you mean people should avoid using an pseudonym in favor of posts that are anonymous, so that there's never any created identity to exploit/defend?
Sorry, it was a bad joke, there's a phrase "don't sign your posts" used when someone ends one with an insult. I support signing your posts with digital signatures if you want.
-----BEGIN PGP SIGNATURE-----
iHUEARYKAB0WIQQC37hdRRO1LtrTQY8AXxvbqjG5KgUCZ5QRXwAKCRAAXxvbqjG5 Kth4AQCccNygglcSyEiMAqQyw6cXH54fnqBT9rJO9TSIqH14rgEAyUwxiQlV05XV Du2ftMk3DwiUZLKDxVI+ODCn4osf2wM= =XZhX -----END PGP SIGNATURE-----
Also there’s something really uncomfortable about the phrasing of a lot of those answers. I mean, even as somebody with an engineering degree, I try not to ever answer a question “as a <field> engineer” because when screwing around online I haven’t done the correct amount of analysis to provide answers “as an engineer” ethically (acknowledging the irony of using the phrase here, but, clearly this is not a technical statement so I think it is fine). The bot doesn’t seem to have this compunction.
This ravenprp guy was an engineering student a couple years ago. I guess it’s less of a thing because he wasn’t commenting under his real name. But it seems like this site, given the type of content it hosts, could easily end up impersonating somebody “as an engineer” in the field they work and have a professional reputation in. And the site even has a historical record of them asking and answering questions through their education, so it does a really good job of misleading people into thinking an engineer is answering their questions.
I know the idea of an individual professional reputation has taken a beating in the modern hyper-corporate world. But the more I think of it, the more I think… this seems incredibly shitty and actually borderline dangerous, right?
It was an early stock discussion forum. It grew rapidly when search engines started indexing everything and this forum had a URL for each message that was easily indexable.
It's still around, but nothing like the old days.
https://www.siliconinvestor.com/subject.aspx?subjectid=36035
Suppose the average post is about 1 paragraph long. One paragraph is about 150 words. So 192211 * 150 = about 29 million words. For comparison, the Lord of the Rings trilogy is only around half a million words.
It wouldn't surprise me if there are more words about Qualcomm in that thread than the total amount of internal and external documentation and financial guidance that Qualcomm itself has ever produced.
Surely users aren't expected to read the entire thread before adding a post? But I think I remember seeing old forums where that basically is the expectation. And honestly... that's pretty cool. It seems better than the new social media, where we keep having low-effort recurring debates. I like the idea of adding to an enormous pile of scholarship in cyberspace. A Ship of Theseus discussion which may outlive any individual participant, but has a semblance of continuity all the same, like an undergraduate college society with a 100+ year history.
Time for a cyberpunk revival. Retro-cyberpunk, we could call it.
I remember visiting those forums when I was young and feeling like part of a big group of friendly people hanging out online together.
I tried creating a new account recently and it had a very different vibe. Felt like the old guard had been established and the forums I looked at were dominated by a couple of posters who just wanted to talk, but not discuss anything.
Some of the post counts of those people were eye-watering.
I think this is the case for most places, I'm afraid. I use mainly Discord - there are certainly a lot of servers where I'm purely because I'm talking to people I met there, and I don't even play that game anymore.
There solution is simpler - after time we create private servers or channels for the old guard, but even then the places deteriorate.
It's a thing I don't know how to solve.
There has to be an active commitment to include (annoying, tactless, socially-impoverished) newbies, or the snake eats its tail and collapses under its own weight.
Is there anywhere on the internet that still has camaraderie?
Other places seem to either not have critical mass to stick around (datatau), or become troll sites (econjobrumors, reddit).
I wonder why there is such a difference.
Such as?