The Karpathy Interview, 6 Months After AI 2027

I find this article quite poorly constructed. It does not even give a statement about the reasoning behind why Karpathy said getting to https://ai-2027.com is unlikely. It also does not clearly define what AI 2027 is?

The following paragraph is almost complete gibberish:

"For AI experts, Karpathy's view is a better counterargument to short timelines than ours. But for non-AI-experts, we think the practical considerations we raised are worth reflecting on with 6 more months of evidence. As forecasters, this is more of an "outside view" - regardless of how exactly AI improves, what problems might slow down an R&D-based takeoff scenario?"

Why would Karpathy's view be different for AI and non-AI-experts?

Did they use AI to write the article?

So much delusion on display here. The folks talking about replacing 95% of remote jobs by 2030 in particular seem to have no idea what people actually do in their work, and how decisions get made, not to mention the actual current state of generative AI.

I hate to lean on credentialism and experience, I really do. But is Karpathy the only one of these who is a) an engineer and also b) over the age of 30?

Why are these relevant? Engineer, because we are talking about a set of technologies that are engineering projects. There is no substitute for hands-on experience in systems. And likely an engineer has taken at least one course that included some history of AI to give a sense of the time scales involved in getting from the perceptron to Sonnet 4.5.

Over 30 primarily because that's roughly old enough to have seen at least one tech hype cycle through which to filter the AI hype cycle. (Some people are old enough to remember the predictions that nobody would use screens in 2025, everything would be a voice interface. Or how economics had fundamentally changed and companies didn't need to make money in the New Economy. Or how Tesla would for sure have 1 million robot axis on the road in 2020. Etc.)

IMHO it's a bearish sign that boosters are not looking to experienced engineers for this kind of analysis.

If you enjoy this kind of trend forecasting, you might also be interested in seeing which colors are expected to dominate clothing stores this winter:

https://www.pantone.com/articles/fashion-color-trend-report/...

Not a snark comment genuinely curious because I’m confused: is Scott Alexander a psychiatrist or clinical psychologist or some kind? Why would he be involved in something like this?

Unless I’m confusing Slate Star Codex and this is a different S.A.

The whole thesis that AGI will come about through advances in coding agents makes zero sense. This isn't a coding problem - it's first a matter of defining the goal ("AGI" means nothing), then considering architectures and learning algorithms, etc, capable of achieving it. What's needed isn't agentic coding ability but rather creativity and ability to design new learning algorithms that are not to be found in the training data.

Coding is the least problem, and I'd guess today's Claude Code, etc is well capable of doing the drudgework.

I honestly thought this was meant as humour… then I realised its intended as serious. Surreal bubble...

I feel like it's important to keep in mind that everyone who has predicted a concrete near-future timeline at all is completely full of shit.