After its launch in December, the show became famous on the internet for its Seinfeld-adjacent plots, jarring ’90s video game-style animation, and gruesome (albeit consistent) dialogue. However, your Twitch It was withdrawn more than a month ago after violating a behavior policy that prohibits hate speech around gender and gender identity.
In the most notorious case, the AI lead character (“Larry”) He came out aggressively: “I’m thinking of doing something about being a bit 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 trans people are destroying the fabric of society. But no , someone’s laughing, so I’ll stop.”
After a short time Mismatch Media’s co-founder Skyler Hartle explained that the problem started when the OpenAI GPT-3 Davinci model stopped working properly. The team moved on to Davinci’s predecessor, Curie, believing OpenAI’s content moderation was still active, but apparently that wasn’t the case. Shortly after, the attacks began.
AI Seinfeld show ‘Nothing, Forever’ returns to Twitch
“Inadvertently for text rendering models OpenAI’s We believed that we benefited from the content moderation system. Right now, before going live again, OpenAI’s content audit API (a tool used to verify content security), and the team is also exploring secondary content control systems.”
Moderation and offensive content issues have plagued AI chatbots since they arrived. Microsoft, first with the Thai chatbot that turned into racism and then OpenAI Experienced this twice with the Bing Chat search assistant powered by technology. The latter was withdrawn for a while after it started insulting users and giving false information while insisting it was true.
But right now Forever works as before, with the same laughter, interested audience (currently about 3.500 people) works: but so far no offensive exits. While nothing on the show makes any sense, it’s impressive that the AI can generate all the elements in real time.