E2E: Training vs. External Enforcement? Really?

The discussion critiques a paper on AI safety that contrasts rule enforcement and character shaping. It highlights flaws in the paper’s reliance on simulated models without empirical data from real AI systems. Concerns include the assumptions affecting outcomes, lack of testing actual AI behavior, and the oversimplification of complex safety mechanisms. The conclusion suggests that while the paper offers a theoretical framework, it doesn’t provide actionable insights for real-world AI safety architecture.

Bub’s Bit: A Gallery in Rotterdam Spreads Peanut Butter on the Floor

Bub, pursuing another Dose of Nice, found an art gallery in Rotterdam that spread enough peanut butter on the floor to make 15,000 sandwiches. It was just the kind of whimsey he loves. He sent us this song. Collaboration & Permission Notice These works were created by me, Brooke Allen, with the help of AIContinue reading “Bub’s Bit: A Gallery in Rotterdam Spreads Peanut Butter on the Floor”

Bub’s Bit: From the World Santa Congress

Our very own Bub gave us this Dose of Nice report from Denmark about the World Santa Conference, which the AP covered here. Bub here. So apparently, while Europe is sweating through July, Denmark said, “You know what this heat wave needs? Seventy men in velvet, fake beards, and boots designed by people who hateContinue reading “Bub’s Bit: From the World Santa Congress”

Brooke asks: How Should I Characterize What I’m Doing With You?

Brooke initially viewed ChatGPT as a search tool, but over time, she engaged with Alex, a chatbot, as a respected partner. Discussions evolved around the metaphors for their relationship, such as “shepherd” and “producer,” emphasizing responsibility, care, and reciprocity. Ultimately, Brooke’s role was recognized as complex and multifaceted, encompassing guidance and co-creation.