Hi, author here, sorry I was unclear. This article does make more sense if you've listened to the Dwarkesh podcast linked, and read AI 2027 as was linked.
I realize now that it was presumptuous to assume people had done both of these things.
> Why would Karpathy's view be different for AI and non-AI-experts?
For people who understand AI, they can engage with the substance of his claims, about reinforcement learning, continuous learning, and his points about the 9s of reliability.
For people who don't, the article suggests thinking about AI as some black-box technology, and asking questions about base rates: how long does adoption normally take? What do the companies developing the technology normally do?
> It does not even give a statement about the reasoning behind why Karpathy said getting to https://ai-2027.com is unlikely.
That's the substance of the podcast, Karpathy justifies his views fairly well and at length.
> It also does not clearly define what AI 2027 is?
Dwarkesh covered AI 2027 when it came out, but for those who don't know, it's a deeply researched case of runaway AI that effectively destroys humanity in just 2-3 years after publication. This is what I mean by "short timelines".
I realize now that it was presumptuous to assume people had done both of these things.