I found this in 2014-15, I forget exactly where, but it was on the public internet.

I remember showing it to some friends on IRC at the time.

It's from the "rump" session that they do at the end of these crypto conferences, where people can do a 5 minute presentation on something that wasn't worthy of a formal presentation in the conference.

It seemed pretty amazing find that video at the time, but looking back, it isn't that interesting (technically! from a history perspective it's really cool), because afaict he was using the basic Cramer-Damgaard ideas about DL-based zkp systems. These are quite badly inefficient in terms of space and computation, for doing hash preimages (and for all they're bad for SHA1, they're way worse for SHA2 of course).

(If anyone is interested, you can watch my 25 minute presentation on "Sigma Protocols" at Advancing Bitcoin 2022 for some background on how you get to such ZKP systems).