The open problem of AI Corrigibility explained by Liron Shapira

In this clip ✂️cut from the must-watch debate between though leaders: Liron Shapira and Roon.

luminary Liron is giving his take on the corrigibility problem:
the crushingly hard open scientific problem of making AI systems that humans can easily fix or change, even if such actions might interfere with their original goals.

Liron Shapira

Liron Shapira is an entrepreneur, angel investor, and has acted as CEO and CTO in various Software startups. A Silicon Valley success story and a father of 3, he has somehow managed in parallel to be a “consistently candid AI doom pointer-outer” (to use his words) and in fact, one of the most influential voices in the AI Safety discourse.

A “contrarian” by nature, his arguments are sharp, to the point, ultra-rational and leaving you satisfied with your conviction that the only realistic exit off the Doom train for now is the “final stop” of pausing training of the next “frontier” models.

He often says that the ideas he represents are not his own and he jokes he is a “stochastic parrot” of other Thinking Giants in the field, but that is him being too humble and in fact he has contributed multiple examples of original thought ( i.e. the Goal Completeness analogy to Turing-Completeness for AGIs, the 3 major evolutionary discontinuities on earth, and more … ).

With his constant efforts to raise awareness for the general public, using his unique no-nonsense, layman-terms style of explaining advanced ideas in a simple way, he has in fact, done more for the future trajectory of events than he will ever know…

In June 2024 he launched an awesome addicting podcast, playfully named “Doom Debates”, which keeps getting better and better, so stay tuned.

Complexity is in the eye of the beholder – by Liron Shapira

AI perceives humans as plants

Rocket Alignment Analogy

Interviews and Talks

Industry Leaders and Notable Public Figures

Lethal Intelligence Microblog

Blow your mind with the latest stories

Favorite Microbloggers

×