Sayash Kapoor

In ordinary research, too.

In 2020, a PhD student at Princeton named Sayash Kapoor was taking a graduate seminar called Limits to Prediction.

He looked at a set of political science papers claiming that machine learning could predict civil wars better than older, simpler statistics.

That's a big claim, predicting civil wars.

We do not think that viewing AI as a human-like intelligence is currently accurate or useful for understanding its societal impacts, nor is it likely to be in our vision of the future.

AI as Normal Technology, Avin Narayanan and Sayash Kapoor, April 15, 2025 End quote.

But I think much of their communication and modeling would be clearer and more accurate if they unbundled their three core frameworks into A description of current AI capabilities, plausibly true at the time, increasingly false.

A prediction of where current AI capabilities are likely to be headed, again, seems false absent a large dispositional shift from the labs or a change in the policy environment that makes untenable their current strategy of automate everything, including and especially safety, and hope it goes well.

Is it fair to say then that y- you're sort of very sympathetic to this, this framing, which is quite influential these days.

It comes from Arvind Narayanan and Sayash Kapoor, that AI is a normal technology.

Or, or the very least, we should have a strong presumption in favor of the view that it's a normal technology.

Whereas where many people working at these frontier AI companies and many people in effective altruist and the rationalist community, where they're coming from is the perspective that, no, it's an extremely abnormal technology, such that the kinds of lessons we might draw from electricity, from standard engineering practices, from the internet and so on, won't necessarily generalize to AI.

How do you actually use the tool sitting on your desk without losing your own mind in the process?

Okay, so if we're gonna start with capability, sorting out what is actually real from what is just, you know, a hallucination of the marketing department, we have to look at Arvind Narayanan and Sayash Kapoor's book, AI Snake Oil.

Yes, absolutely essential reading.

Because AI isn't just a single technology.

End quote.

Sayash Kapoor reports that ultra-fast is a difference in kind.

You only get an effective two times for time speed up because you can't speed up tool calls, but there is a big shift because if AI is fast enough you don't have to break flow or context shift.

It's also a big win for computer use.

To access it in Codex and ChatGPT Work, we're introducing Pro five hundred, a new plan with our highest usage limits, twenty-five x plus, and access to Ultra Fast.

Sayash Kapoor reports that Ultra Fast is a difference in kind.

You only get an effective two x to four x speed up because you can't speed up tool calls.

But there is a big shift because if AI is fast enough, you don't have to break flow or context shift.