168 - How to Solve AI Alignment with Paul Christiano
Inside the episode
Paul previously ran the language model alignment team at OpenAI, the creators of ChatGPT.
Today, we’re hoping to explore the solution-landscape to the AI Alignment problem, and hoping Paul can guide us on that journey.
In today’s episode, Paul answers many questions, but the overarching ones are:
1) How BIG is the AI Alignment problem?
2) How HARD is the AI Alighment problem?
3) How SOLVABLE is the AI Alignment problem?
Does humanity have a chance? Tune in to hear Paul’s thoughts.
9:20 Percentage Likelihood of Death by AI
19:15 Chimps to Human Jump
21:55 Thoughts on ChatGPT
27:51 LLMs & AGI
32:49 Time to React?
38:29 AI Takeover
41:51 AI Agency
51:14 Training AIs to Be Honest
59:36 How Solvable Is the AI Alignment Problem?
1:03:48 The Technical Solutions (Scalable Oversight)
1:16:14 Training AIs to be Bad?!
1:18:22 More Solutions
1:21:36 Stabby AIs
1:26:03 Public vs. Private (Lab) AIs
1:28:31 Inside Neural Nets
1:32:11 4th Solution
1:35:00 Manpower & Funding
1:38:15 Pause AI?
1:43:29 Resources & Education on AI Safety
1:49:00 Paul’s Day Job
1:50:15 Nobel Prize
1:52:35 Treating AIs with Respect
1:53:41 Uptopia Scenario
1:55:50 Closing & Disclaimers
Alignment Research Center
Paul Christiano’s Website