AI Links, 12/06/2025

Andrey Mir on AI and epistemology; Andrew Yang on AI and the job market; Timothy Taylor on business usage rates; Dwarkesh Patel on AI shortcomings

Andrey Mir writes,

At some point, AI-generated content will exceed the amount of content produced by humankind in our entire history. Not only will the human share decrease, but the AI slop aimed exclusively at grabbing attention will also increase, polluting the joint AI–human “knowledge.” New technologies of epistemic vigilance will certainly be needed.

He is meditating on Generative Knowledge, by Paolo Granata.

And what if AI’s advantage in knowledge-generation becomes qualitative?

It is conceivable that AI does not simply flood the zone with slop, but becomes better at coming up with new scientific discoveries, new management techniques, new military strategies, and new artistic creations. You can assert that this sort of qualitative superiority is impossible, but perhaps that is just your carbon bias showing.

Mir points out that considering AI as a potentially superior knowledge-generator actually opens up deep questions of epistemology.

Andrew Yang writes,

A partner at a law firm told me, “AI is now doing work that used to be done by 1^st to 3^rd year associates. AI can generate a motion in an hour that might take an associate a week. And the work is better. Maybe I have to check the work or compare multiple models, but that might take me an hour or two. Someone should tell the folks applying to law school right now.” Also this year, law school applications surged 21% - there’s a flight to safety, though in this case it’s not so safe. Three years from now, how many graduates are going to be hired?

…We’re not setting our young people up for success at all. I wish it were different.

Timothy Taylor writes,

Here’s a figure showing businesses that are using AI tools: the blue line is those that have used such tools in the last two weeks: this has risen from 5% of firms at the start of 2024 to about 10% here near the end of 2025. The orange line shows firms that expect to use AI tools in the next six months. The lines are trending up, but not skyrocketing.

…Sometimes when a remarkable new technology is being adopted, there is a period of several years where many people are learning about the technology and multiple application are being developed, but this doesn’t show up in the big-picture statistics. Then at some point, some of those applications gain traction with a broad group of users and the technology takes off. But at least so far, with the current abilities of AI and the AI applications currently available to US firms, that launching point hasn’t happened yet.

I expect that to change. I was just at a nearby Firestone service center getting an oil change. An experienced clerk was training a new clerk on using their software. It was painful for me to watch.

Another painful experience was trying to get a question answered from my bank. first, I had to get through all of the menus in their voice response system in order to reach a person. Then the person could not answer my question, because her system was too inflexible.

I think that the user interface for every computer system that currently exists could be replaced by something better that relies on AI. Right now, this vision seems outlandish. I expect that within two years it will take a single prompt to do it. “Claude, here is our consumer banking application. Users struggle to make it do what they want. Fix it.” And within an hour or less, you will have a system where the user just says what they want to do, in English, and the system does everything a teller, or even a manager, would do today. And more.

Dwarkesh Patel writes,

Sometimes people will say that the reason that AIs aren’t more widely deployed across firms and already providing lots of value (outside of coding) is that technology takes a long time to diffuse. I think this is cope. People are using this cope to gloss over the fact that these models just lack the capabilities necessary for broad economic value.

…Models keep getting more impressive at the rate the short timelines people predict, but more useful at the rate the long timelines people predict.

substacks referenced above:

@

Thoughts on AI progress (Dec 2025)

@

Generating knowledge with AI: Epistemic partnership?

@

Pyramids to Columns