90/30 NYC: Last AI Built by Humans

​Paper

What happens when AI can improve itself, including the way it learns to improve?

​Join 90/30 for a reading group on The Last AI Built by Humans: Toward Genuine Recursive Self-Improvement. This paper explores what it would take for AI systems to identify their own limitations, test and retain improvements, and eventually improve the process that produces their next version.

​We’ll discuss:

• ​What separates genuine recursive self-improvement from an AI simply revising its answer?

• ​How much of the improvement process can AI take over from humans?

• ​How do we tell whether a system is getting more capable or just getting better at gaming its evaluations?

• ​What could self-improving systems look like in software engineering, robotics, and scientific research?

​This is a long read, so come with a section, question, or argument you want to unpack. You don’t need to understand every technical detail to join the conversation.

​Event Schedule:

​7pm–8pm: Quiet reading time, grab a snack and read! (optional)

​8pm–9pm: Group discussion about the paper 📝

​9pm–10pm: We have our space for a bit longer, stay to socialize or network!