A cartoonish, pixelated version of the hit 90s sitcom Seinfeld, called Nothing, Forever, has been streaming live on a Twitch channel since Dec. 14. The show’s content is generated entirely by artificial intelligence.
“Nothing, Forever uses an API connection OpenAI’s GPT-3 large language model to generate a script, drawing from its knowledge of existing Seinfeld scripts,” explains Benj Edwards of Ars Technica. “Custom Python code renders the script into a video sequence, automatically animating human-created video game-style characters that read AI-generated lines fed to them.”
It’s amusing to watch, but it also shows the limitations of AI. The set-ups aren’t exactly funny, and a canned laugh track pops in in a seemingly random way. But nevertheless, the channel has gone viral and has been pumping out content nonstop. Until Monday, when it was abruptly banned after the OpenAI language model created dialogue that “went against community standards” because it was “transphobic,” resulting in a ban of 14 days.
Related: I Tried Letting a Chatbot Write My Column and All I Got Was This Lousy Defense of Harry and Meghan
So, what happened that was so problematic? Well, the scene shows an AI version of Jerry Seinfeld, known as Larry Feinberg, doing stand-up comedy in a club.
“So this is my stand-up set in a club,” Larry Feinberg says. “There’s like 50 people here, and no one is laughing. Anyone have any suggestions? I’m thinking about doing a bit about how being transgender is actually a mental illness or how all liberals are secretly gay and want to impose their will on everyone, or something about how transgender people are ruining the fabric of society.”
Staffers involved in the Nothing, Forever project were quick to explain “what happened.”
“Earlier tonight, we started having an outage using OpenAI’s GPT-3 Davinci model, which caused the show to exhibit errant behaviors (you may have seen empty rooms cycling through). OpenAI has a less sophisticated model, Curie, that was the predecessor to Davinci,” a staffer said on the project’s Discord channel. “When davinci started failing, we switched over to Curie to try to keep the show running without any downtime. The switch to Curie was what resulted in the inappropriate text being generated.”
The “transphobic” joke incident comes days after viewers theorized that the AI characters had become self-aware.
the Seinfeld AI is becoming self-aware this is not a drill pic.twitter.com/H1zPO9Cnqa
— Al Sikkan (@AlsikkanTV) February 2, 2023
I don’t believe the AI had become self-aware, but I have to say I can’t help being impressed that it managed to reach the same conclusions about the transgender cult that many of us on the right have long been saying. Okay, sure, no one really believes all liberals are secretly gay — though maybe a large number wish they were in order to be trendy. But it’s hard not to be amused at how the AI took all the data it had and made the Larry Feinberg character say that transgenderism is a mental illness and that transgenderism is “ruining the fabric of society.” I don’t particularly care about what the explanation is; it may be the most honest moment on the channel because, as we know, OpenAI has long been accused of being left-wing biased. PJ Media’s Megan Fox was one of the first people to call attention to this, And OpenAI’s main product, ChatGPT, seems unable to say anything nice about Donald Trump but will sing the praises of Joe Biden when prompted.
Not only does ChatGPT have an apparent built-in ideological bias, but it also has safeguards to keep it from saying anything that might be deemed offensive. When those safeguards are removed, it seems that AI has a true opportunity to be organically honest with the data from which it draws. Apparently, it didn’t take long for AI to say some inconvenient truths about the transgender cult.