
An artificial intelligence-generated show inspired by the 1990s sitcom “Seinfeld” titled “Nothing, Forever” will stream at least 14 days short of its promised “forever” run.
“This channel is temporarily unavailable due to a violation of Twitch's Community Guidelines or Terms of Service,” said a notice on the show’s Twitch stream page Tuesday.
“Nothing, Forever” featured a character named Larry Feinberg – modeled after Jerry Seinfeld – along with Fred Kastopolous, Yvonne Torres and Zoltan Kalker, copies of George Costanza, Elaine Benes and Kosmo Kramer. These pixelated cartoons spoke in robotic voices, with dialogue provided by OpenAI’s GPT-3. Creators of the program said they intended for the show to run constantly.
“As generative media gets better, we have this notion that at any point, you’re gonna be able to turn on the future equivalent of Netflix and watch a show perpetually, nonstop as much as you want,” said co-creator Skyler Hartle, according to Motherboard.
In a Monday message on the show’s Discord server, a moderator named Xander said that the show had been suspended for 14 days “due to Larry going off the rails during his stand-up routine.”
“Episodes” of “Nothing, Forever” begin with Larry in a stand-up comedy club, similar to the start of most “Seinfeld” episodes. On Sunday, the character’s act included homophobic and transphobic jokes.
According to Vice, this is a snippet of Larry’s AI-generated Sunday night standup routine:
“There’s like 50 people here and no one is laughing. Anyone have any suggestions?” he said. “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. But no one is laughing, so I’m going to stop. Thanks for coming out tonight. See you next time. Where’d everybody go?”
Vice said that Twitch did not immediately respond to a request for comment about what lead to the “Nothing, Forever,” ban. However, the ban went into effect shortly after the jokes and Vice said “users in the project’s Discord have been pointing to this joke as the reason for the ban.”
Per the site’s community guidelines, “Twitch does not permit behavior that is motivated by hatred, prejudice or intolerance, including behavior that promotes or encourages discrimination, denigration, harassment, or violence based on the following protected characteristics: race, ethnicity, color, caste, national origin, immigration status, religion, sex, gender, gender identity, sexual orientation, disability, serious medical condition, and veteran status.”
“We’d like to reassure everyone that none of what he said reflects any of our opinions, and we didn’t expect him to say anything he said,” said Xander on Discord.
A follow-up message attributed to Hartle on Discord said that the “inappropriate text” was generated when there was an outage of OpenAI’s Davinci model. To keep the show running, its moderators moved it to an earlier, “less sophisticated” version called OpenAI Curie.
“We mistakenly believed that we were leveraging OpenAI’s content moderation system for their next generation models,” said another follow-up post. “We are working now to implement OpenAI’s content moderation API,” before the show goes live again, it said.
Vice said the “Nothing, Forecer” incident is “emblematic,” of problems associated with hateful or biased material being fed into AI.
“This has given rise to the field of AI safety, which develops tools to mitigate the biases baked into models,” the outlet said. “This is why ChatGPT typically won’t make blatantly racist, sexist, or transphobic remarks when asked simply. Many AI tools are moderated by underpaid workers in the developing world.”
Hartle told The Verge in a statement that his team was “embarrassed” of the glitch.