Jared Kaplan

It really was.

I mean, you had Amodei, his sister Daniela, and Jared Kaplan, who is now the chief scientist at Anthropic.

And they were applying these strict probability models to global catastrophes.

Exactly.

Roone is emphasizing mechanistic interpretability as the way.

Anthropic's new responsible scaling officer is once again Sam McCandlish, replacing Jared Kaplan.

Heading.

Open weight models are unsafe and nothing can fix this.

I mean, it, it helps to know, I think, that Dario Amodei is a sort of neurotic by temperament, um, but he is also part of this community of AI safety worriers in the Bay Area.

You know, at one point in 2016, he's, like, living in a group house with Paul Christiano and Chris Olah and Jared Kaplan and sort of all these now, like, leading lights of AI and AI safety, and they are all just sitting around at night, like, reading posts from Lesswrong, you know, and other, like, AI safety-focused publications and talking about them and debating what the sort of existential threats from AI might be.

And this was considered insane behavior in 2016 [laughs] because nothing actually worked yet.

But they were sort of looking, peering into the future and imagining when systems did work, um, and could be improved through scaling, what that would mean for safety.

signed by 1,386 employees of different frontier labs.

OpenAI's chief scientist, anthropics Dario Amode and Jared Kaplan were also part of some of the conversation around this.

He had some signatures from DeepMind, and all these signatures were basically asking the U.S. government to help, quote, build, and this is asking the government, quote, the technical and governance tools needed to deliberately pace the frontier.

So, again, they're asking for stuff to regulate themselves.

We'll start in January with the OpenAI paper called the Neural Scaling Laws.

Lead author is Jared Kaplan.

There's a whole bunch of others, including Daryl Modi, who this time is at OpenAI, will eventually be the CEO of Anthropic.

And this was the paper that sort of kicked off this idea of, you know, different from all the other machine learning models we did where we worry about overfitting and training sets and so forth, that for large language models, we should just train on as much data as we can possibly get.

But towards the end of my PhD, I became convinced that working in scattering amplitudes was not, um, the, the most profitable thing I could do in terms of making an impact on, on the sum total of human knowledge.

Um, funnily enough, many of the, the people that I was citing at the time of my PhD, people like Jared Kaplan and Geoffrey Pennington also made a similar decision, um, around about the same time.

So in, uh, 2016, towards the end of my PhD, I saw the AlphaGo match.

I remember watching that, um, like so many other people, being fascinated by Move 37, and I became convinced that the future was going to involve agents that could really aid humans in making discovery and accelerate the rate of scientific progress.