OpenAI's LLM spec and Superalignment AGI team

Two AI's walk into a bar. One of them says to the bartender, 101101, and the other says, 1110. 🤖🤣 There was so much AI activity last week that there were two stories that went way under the radar that I want to make sure to highlight today. The first was that OpenAI's super alignment AI safety team, which is a mouthful, don't get me wrong, was disbanded. And what that means for us normal humans. There's no general oversight anymore on AGI and how it affects us as human beings. This is something that people honestly should be a little freaked out about. I'm not really in the AGI camp, to be honest. I think there's some quantum limitations to it. But there's so much potential for things to go wrong if there's a lack of supervision that I think it was a pretty big mistake to disband this team. “I lost trust”: Why the OpenAI team in charge of safeguarding humanity imploded Company insiders explain why safety-conscious employees are leaving. by Sigal Samuel at Vox (18 May, 2024) https://www.vox.com/future-perfect/2024/5/17/24158403/openai-resignations-ai-safety-ilya-sutskever-jan-leike-artificial-intelligence The other thing that flew under the radar. Last week, OpenAI released a specification in an attempt to self regulate kind of what like the bitcoin and cryptocurrency industry did. They're trying to make a specification document that will allow anyone to be able to describe what the expectations of their models are. What are the things that it should say? What are some of the things that it shouldn't say? https://openai.com/index/introducing-the-model-spec/ It is kind of interesting that they're trying to get ahead of the regulations. Time will tell if this actually prevents any formal legislation coming out, but honestly, if you're in the business of rolling your own LLM, you're sort of in a special category all on yourself. And I'm not sure if the specification is really going to take hold. #generativeai 🎧 Music by V - "FRI(END)S"