OpenAI CEO Sam Altman has noted that social media users are increasingly adopting the writing styles of large language models (LLMs), a trend he observed while reviewing forum posts about Codex, OpenAI’s agent tool for developers.
Altman described the experience as “strangest” while going through the posts, noting that many seemed potentially fake or bot-generated. However, he confirmed that the underlying trend of strong Codex growth was genuine despite the suspicious content. He suggested several potential explanations for the proliferation of LLM-like content, including the possibility that “real people have picked up quirks of LLM-speak,” thereby mimicking the language patterns of AI models.
Altman also pointed to the tendency of the “Extremely Online crowd” to converge on similar communication styles as a contributing factor. Additionally, he cited the hype cycle surrounding AI tools, characterized by extreme reactions, and “optimization pressure from social platforms on juicing engagement” as other potential causes. The overall effect, according to Altman, makes “AI twitter/AI reddit feel very fake” compared to the online environment of just one or two years ago.
This sentiment echoes a previous post by Altman where he noted a significant number of “LLM-run twitter accounts,” leading him to reconsider the “dead internet theory.” Paul Graham, co-founder of Y Combinator, responded to Altman’s observations, confirming that he too has noticed the same trend on X. Graham highlighted that AI-generated content is not limited to “fake accounts run by groups and countries that want to influence public opinion” but is also being produced by “a lot of individual would-be influencers.”
The concerns about the rise of low-quality, AI-generated content, often referred to as “AI slop,” extend beyond Altman and Graham. Substack CEO Chris Best addressed the issue on “The a16z Podcast,” warning that “sophisticated AI goon bots” will flood the media with engagement bait. Best predicted a future where an abundance of “AI slop” will keep users mindlessly clicking, potentially degrading the quality of online discourse and information.
