Thomas Hickey is a Dutch student in both philosophy and artificial intelligence at Utrecht University. We talk about his bachelor thesis on A.G.I. or Artificial General Intelligence, on bottlenecks for recursive self-improvement. But also go into existential risk and what we can do about it.
Alex van der Meer
email:
[email protected] but remove capital letters, (for spam)
https://twitter.com/AlexvanderMeer5
https://www.youtube.com/@safetransition9743
Eliezer Yudkowsky, on why it is lethal not to have retries
https://intelligence.org/2022/06/10/agi-ruin/
OpenAI on their approach to alignment blog post
https://openai.com/blog/our-approach-to-alignment-research/ On some podcast players you should be able to click the timestamp to jump to that time.
(03:16) - Recursive self-improvement, how long until superintelligence
(10:50) - What can be learned in the digital realm
(14:21) - How fast can it learn in the real world
(18:34) - Can AGI become better than us?
(22:54) - Complex enough environment to create superintelligence?
(29:10) - Can AGI Thomas take over the world?
(37:40) - Is superintelligence irrelevant for safety?
(41:38) - Existential risk from AI?
(48:09) - How to decrease the chance of a bad outcome?
(53:19) - ChatGPT and the best current models
(59:57) - Solution to the treacherous turn?
(1:05:01) - AGI becomes religious?
(1:11:03) - Starting point of the intelligence explosion?
(1:16:49) - OpenAI Alignment approach blog post
(1:18:29) - Is Open source bad for safety?
(1:24:49) - How to contact me
