Agreed - most of the impact here will be from domain experts applying it in niche areas.

The people hyping it generally have no domain knowledge in anything.

The problem is that jumping on cop cars alienates the median voter.

MN's protesters won over voters this time; that's why change is happening.

If you want change, you have to change opinion in the middle, and they did an amazing job here.

You're right. Self-selection is the more correct phrase than adverse selection to explain the dose response.

As you note elsewhere, it looks like the researchers agree that this is a limitation, and all we can really say is that there's a correlation of depressed people using chatbots more.

The alternative explanation (that the dose response just shows adverse selection) is also a much simpler mechanic. Of course people who lack high trust relationships will be more likely to turn to a chatbot.

Dismissing it with a sarcastic "cool explanation bro" doesn't make for good epistemics.

Vulnerable people seeking it out being bad is an assumption.

What's the counterfactual? If it's always immediate access to a professional therapist, then sure, it's worse. But what a counterfactual.

I tested ChatGPT in private mode with a suicidal note, and got immediate support hotlines.

What is bad about the building the city?

If you had to choose an outcome between 50% of images with no alt text, or 99% with LLM alt text, which would you believe to be better?

This is because of automation, not globalism. Manufacturing is still a major part of the economy.

In fact, when you look at actual units of goods produced, manufacturing’s share of GDP has been stable for half a century.

www.stlouisfed.org/on-the-econo...