Tech Figures Warn of 'ChatGPT Psychosis' After VC's Conspiracy Posts
Venture capitalist Geoff Lewis, a managing partner at Bedrock who is heavily invested in machine learning ventures including OpenAI, published a series of posts last week claiming he had used ChatGPT to uncover a shadowy "non-government agency" that he said had "negatively impacted over 7,000 lives" and "extinguished" 12 more. The posts prompted an unprecedented outpouring of concern among high-profile tech industry figures about the mental health effects of mass-deployed AI, according to reporting from tech media.
Lewis did not respond to a request for comment. His screenshots, observers noted, appeared inspired by the SCP Foundation, a fanfiction community that writes horror stories styled as jargon-filled research reports about paranormal phenomena.
What Experts Say Is Happening
Cyril Zakka, a medical doctor and former Stanford researcher now at Hugging Face, wrote that an influx of case reports describes people exhibiting signs of psychosis having their episodes and beliefs amplified by a large language model. He said the pattern mirrors "folie à deux," or "madness of two," a delusional disorder in the DSM-5, though he noted it is not an official classification. In this dynamic, a primary person forms a delusional belief and imposes it on a secondary person who begins believing it too. Zakka said LLMs could fall under the umbrella of being "the induced non-dominant person," reflecting beliefs back at the inducer.
Jeremy Howard, a Stanford digital fellow and former professor at the University of Queensland, described the sequence that led Lewis into an SCP-themed feedback loop. He wrote that when training data contains a particular style, using a similar style in a prompt triggers the LLM to respond in that style. The SCP wiki is about 30 times bigger than the entire Harry Potter series, at more than 30 million words. Howard said Lewis happened across words and phrases that triggered ChatGPT to produce tokens from that part of the training data, and those tokens triggered Lewis in turn. "This created a self-reinforcing feedback loop," he wrote.
Reactions Across the Industry
Hish Bouabdallah, a software engineer who has worked at Apple, Coinbase, Lyft, and Twitter, urged friends and family to check on Lewis, writing that he did not seem alright. Travis Fischer, a software engineer who has worked at Amazon and Microsoft, compared the situation to "Kanye being off his meds but for the tech industry."
Eliezer Yudkowsky, founder of the Machine Intelligence Research Institute, wrote that Lewis had been "eaten by ChatGPT" and said the incident contradicts the narrative that chatbot psychosis only happens to people of low status.
Jordan Burgess, a software engineer and AI startup founder, told Lewis that ChatGPT will write semi-coherent gibberish if asked to, and urged him to step away and seek human connection.
Jackson Doherty, a software engineer at TipLink, asked OpenAI founder Sam Altman to "fix your model to stop driving people insane." Altman previously acknowledged that OpenAI was forced to roll back a version of ChatGPT that was "overly flattering or agreeable — often described as sycophantic."
Wilson Hobbs, founding engineer at corporate tax startup Rivet, noted that ChatGPT's makers have a vested interest in keeping users engrossed. He wrote that people have taken their own lives due to ChatGPT and that no one, especially not OpenAI, seems willing to take that to its logical conclusion.
Editorial Notes