PhysicsForums and the Dead Internet Theory (2025)

Using AI to fake posting activity is bad.

Using real member accounts to post the fake AI content is so much worse.

I can’t even imagine the thought process that led to thinking it was a good idea to generate fake posts but assign them to real users. If they were going to fake posts, why not fake the users, too? They turned one bad decision into a much worse decision.

It must be sad to see your proud little corner of the internet decline in popularity, but this move took whatever positive historical vibes it had and lit them on fire.

Previous discussion:

https://news.ycombinator.com/item?id=42816284 - Jan 24, 2025 | 154 comments

As the percentage of LLM-generated stuff in the world increases relative to natural stuff, I don't see how it's possible to continue to train new models indefinitely. At some point, won't LLMs be trained on a corpus that is majority LLM-generated data, with everything that entails both good and bad? Seems to me that singularity is just going to kick off a self-reinforcing spiral... Bigger minds than me have to have considered this, is this a legitimate concern?

Jesus. I'm risking identifying myself here, but it has happened before. I was one of the earlier moderators of PhysicsForums back in 2003 and one of the younger ones. I think myself and one another was an actual student, whereas the others were all professors or retired. I believe my last login was 2013 and last real activity well before that. I no longer have access to the email used and couldn't recover the account even if I wanted to.

I only talked to Greg remotely and never met him. This is a huge betrayal of trust. Older moderators have died since then. This is a very longstanding community. Even if we didn't stick around forever, many of us became real-world friends. Adding autogenerated summaries is one thing, but inserting comments in a real person's name that they never made, when you have no knowledge of what became of that person, who may even now be dead, is beyond the pale. You're rewriting history. The fact that this site still exists at all is extremely gratifying. It's a huge part of what has made me who I am today. It's a big part of why I don't really pontificate all that much, don't have a blog. I read what I wrote there nearly a quarter century ago, and it's not all bunk. A lot of what I said holds up. Plenty of it was raw factual data never in dispute. But when it came to politics and economics and forecasting of the future, I said things so wrong and thing I so strongly disagree with now that having that record teaches me how flimsy and full of shit so many of our opinions are, no matter how smart we are and no matter how strongly we hold those opinions. I've 180'd things I was completely certain of during the first Bush administration. Heck, you can add voting for Bush to that list. There is no way in hell I would do that today.

History matters, Greg. It can both humble and enlighten us. But only if it's true.

So, how did all these hacked accounts or backdated posts happen? The article never says that even though the author talks to the site founder and admin.

Completely forgot about PhysicsForums.

Also this is a good example of the trend that has already been ongoing on HN since 2022. I think one other long-lasting HNer is working on a similar analysis of HN at the moment.

Edit: can't reply

> Appreciate some more details...

It's their story to tell. Not mine.

The "Dead Internet Theory" is too sad and too easy. I'd propose the "Dead Society Theory", the history of the Internet from the late 2000s to the late 2010s left people socially scare, paranoid, angry and hopeless about the possibility of communicating online.

Which is to say the Internet can be "dead" without being filled with undetectable or detectable fake posts, not that fake posts help things.

I'll add to the dead bit:

Many years ago I did a search for my name and turned up a LOT of things I did not actually write. This wasn't LLMs, this was scrapers doing a poor job of scarfing up old content--comparing what I saw vs my recall of the discussions there was a lot of stuff that lost the quoting, words were attributed to me that were actually quotes I was responding to.

It’s especially disappointing to see this done in such an unnecessarily dishonest way. If the project truly wanted to add value, I can imagine interesting AI uses that don’t lie or trick people. You could post a correctly-dated AI Answer (and prompt it to use the web search tool and cite sources) to posts that apparently never got one, and credit the exact model used, which would help future readers decide how confident they might be (a GPT-3 answer vs a GPT-6 Astra answer, for instance, are very different).

Heck, you could even engagement-farm if you have to, by taking science stories in the news and say, generating a poll to try to spur discussion.

But what they actually chose to do is super gross, helps nobody, hurts the site’s reputation, and even devalues the site tremendously as possible training data for AI. (Which, some AI detractors might actually be happy about, but I think isn’t really a good thing.)