Old talks, 2000 - 2020 period.

A list of talks that happened in 2000 - 2020 period, for the future people to just look up or pop up in search results. The links are collected at the start of 2026. Many of those links took a lot of effort to dig up, but then I abandoned this writeup and just posting it now to preserve it I guess. It would be nice of you to point some AI (or yourself) at this to restructure it and assemble more comprehensive archive, with independently stored videos also. But please pass it forward and find a way to post it too. Singularity Summits https://intelligence.org/singularitysummit/ • A lot of link rotted links, but not to majority of vimeo recordings thankfully.• All of the Summit 2012 linkrotted >:(https://nickbostrom.com/media.pdf • Nick Bostrom lists 504 interviews on 1992 - 2012, 493 over 13 of those years, one interview per 13 days. That's a lot of interviews.Hard AI Future Salon, Eliezer Yudkowsky 2006 https://archive.org/details/FutureSalon_02_2006 6/10 Very good actually, and the questions were too, good job secondary characters. Some questions I was not there to ask: • If you paint like 100 digit random number on Corolla, does it become a narrower target? Do you make a distinction of narrowness and utility?• It's really weird that population size doesn't mater for information gain? Obviously if you have larger pop in evolutionary algorithm, you get higher reward? I should probably just check out the paper referenced.• People do copy parts of the companies, approaches, methods that proved to work in real world. So, in some sense there is similar mechanism that makes random picks, and argmax picks, iteratively and in different proportions. Maybe humans are just a better substrate for evolution, so that the same (few) bits of ground truth obtained from failed / successful tests, get utilized better.Eliezer Yudkowsky: The Challenge of Friendly AI 2007 1/10 https://www.youtube.com/watch?v=nkB1e-JCgmY https://youtu.be/b_4fCIoUnNs https://www.youtube.com/watch?v=Zfj