Peter McCrory

of our work.

How do you establish credibility as a research group within Anthropic?

One way that we do that is by publishing work out in the open.

So we put out this labor impact report in early March where we introduced a method for detecting displacement should it materialize.

And part of the reason that we published that null finding

was that it in some ways ties our hands to analyzing the data in a particular way that we think is sensible, that we can revisit over time and assess, is there any evidence yet of displacement or not?

And in the academic literature, there's all this worry about publications that are, you're doing analysis and you're only able to publish analyses that have a finding.

One aspect of our work is we're able to publish work that doesn't have a finding, but can nevertheless lend some useful insight on this question.

So I think here that we should be a bit circumspect about the range of uncertainty before us.

This is the most consequential technology of the last hundred years.

It's a general purpose technology.

Every sector, almost every occupation, as I've mentioned,

is set to be affected.

Adoption has been very fast.

The models themselves are improving very quickly.

At present, the most performant model can handle roughly 14 hours of autonomous work reliably.

And the time horizon that these models can handle is doubling roughly every four to seven months.

That alone would raise questions about how this sort of broad-based automation of cognitive work might very well affect large swaths of the labor market.

You add to the fact that it may very well represent an innovation in the method of innovation.

So macroeconomists talk about the importance of innovation for producing long-run growth.