I've seen some plainly wrong takes on this story along the lines of “this is inevitable, #AI is too dumb to avoid transphobia / other hate."
But it’s not. The scene resulted from fallback to a more rudimentary model less amenable to #OpenAI's moderation tools. Encouraging that in more advanced models this is largely solved & the #watchmeforever team has learned not to fall back to the less moderated model.
From: @arstechnica
https://geeknews.chat/@arstechnica/109825036900248884
Ars Technica (@[email protected])
Attached: 1 image Nothing, Forever's endless AI-generated Seinfeld episode has come to an abrupt halt after language model tools from OpenAI went haywire and generated transphobic content. https://arstechnica.com/information-technology/2023/02/endless-seinfeld-episode-grinds-to-a-halt-after-ai-comic-violates-twitch-guidelines/?utm_brand=ars #bot Original tweet : https://nitter.it/arstechnica/status/1622960157104955392

