Home | Search

2026-08-10

Why my P(doom) is less than Yudkowsky

Disclaimer

Main

Short version of my disagreement with Yudkowsky:
- I think pausing AI research (atleast for some years) via persuading people is easier than Yudkowsky thinks. As AI keeps progressing, it becomes easier in many ways to persuade people. There will be more concrete evidence that ASI could be coming soon. There will be more concrete evidence from toy demos of misalignment. The political leaders of the anti-ASI movement will get more attempts at the hard problem of persuasion, and with each attempt they will become more skilled.
- I think pausing AI research (atleast for some years) via blowing up TSMC is easier than Yudkowsky thinks. You need to only convince one powerful actor (either a state actor or non-state actor) to blow TSMC up and accept the risk of being retaliated against. It is fine if hundred other actors are not convinced, you only need one.
- I think using AI to overthrow the govts of the world and establish your own global dictatorship is easier than Yudkowsky thinks. I still have some chance you can do this using a AI that is powerful but yet within your control. Hence one of the AI companies could themselves takeover the world and then slowdown the AI race.
In short the overarching theme here is I think Yudkowsky understands technical alignment research better than me, but I understand politics better than Yudkowsky. I'm not skilled enough to be an expert politician, but even at my current noob level, I think I have better political intuitions than him.

Enter email or phone number to subscribe. You will receive atmost one update per month.

Alternatively, you can subscribe via RSS, or a third-party service like blogtrottr

Enter comment