X enthusiast and Reddit shareholder Sam Altman had an epiphany on Monday. The bots made it impossible to determine whether social media posts were actually written by humans, he posted.
The realization came when I was reading (and sharing) several posts from R/ClaudeCode SubredDit, which praised Openai Codex. Openai launched a software programming service in May that took over Claude code for Anthropic.
Recently, its subreddit announced that posts from self-proclaimed code users have been so filled that it has moved to a codex where one Reddit user joked.
This made Altman wonder how many of those posts were from real people. “I had the strangest experience reading this. In this case, I know that the growth of codex is really strong and the trend here is real, but I think it’s all fake/bot,” he confessed in X.
He then analyzed his reasoning live-action. “I think a lot is happening. Actual people pick up the quirks of LLM speaking and the very online crowd drifts together in a very correlated way. Bots.”
To decode it a bit, he accusssssssssssssssssssssings LLM of starting to sound like LLM, even though LLM was invented to mimic EM dashes to literally mimic human communication. And Openai’s model was undoubtedly trained on Reddit, where Altman was a board member until 2022, and was disclosed as a large shareholder at the company’s IPO last year.
He states a very useful point that fandoms led by social media users on very, always on tend to behave in strange ways. Many groups can be left to a festival of hatred when overrun by people who vent their frustration with their brothers.
TechCrunch Events
San Francisco
|
October 27th-29th, 2025
Altman also unearths incentives for social media sites and creators to rely on engagement to make money. fair enough.
But then Altman confesses that one of the reasons he thinks this subreddit’s pro-openy post is a bot is because Openai is “Astroturfed”. It usually involves posts made by a person or bot paid by a competitor, or posts paid by a three-time contractor, giving competitors a plausible negativity.
There is no evidence of Astroturfing (although it is possible). However, after releasing GPT 5.0, we saw how Openai Subreddits turned on the company. Instead of a wave of praise from loyal people to the new model, many angry posts were voted. People took them to Reddit and X and complained about everything from GPT personality to how they burned through credits without completing tasks.
The day after the bumpy release, Altman held a Reddit Ask-Me-Anything session on R/GPT, sueing the issue and confessing that he had promised a change. GPT SubredDit has never fully recovered previous levels of love, and regularly posts about how much users hate GPT 5.0 changes. Are they human? Or is it somehow fake, as Altman appears to be implying?
Altman said, “I somehow feel that the net effect is AI Twitter/AI Reddit is very fake in a way that wasn’t actually a year or two ago.”
If that’s true, who is it? GPT has become so good at writing that LLM has become a plague not only for social media sites (there are always bot problems), but also for schools, journalism and courts.
I don’t know if the Reddit posts are written by bots or if humans are fictional accounts using LLMS, but it can be quite a few. Data security company Imperva reported that more than half of all internet traffic in 2024 was primarily non-human by LLMS. Botglock of X said: “The exact figures have not been published, but estimates for 2024 suggest hundreds of millions of bots in X.”
Several cynicians suggest that Altman’s lament is his first foray into marketing Openai’s rumoured social media platform. In April, Verge reported that projects like this were at the earliest stages of taking on X and Facebook. This product may or may not be present. Altman may have an impure motive to suggest that social media has been fake these days.
But, motive aside, if Openai plans to have a social network, what is the chance that it will be a bot-free zone? And what’s interesting is that if it goes viral and bans humans, the outcome would not be different. Not only does LLM still hallucinate the facts, but when researchers at the University of Amsterdam built a fully constructed social network of bots, they soon discovered that bots formed creeks and echo chambers for themselves.
