This is the tough paradox when people claim AI is useful or useless. Almost no one who thinks AI is useless has spend a few hundred hours to learn how to use it, or at least spent that time over six months ago when the models weren't as capable. I basically feel anyone who hasn't used AI for at least 100 hours in the past 3 months (~8 hours per week) has no idea how capable AI is right now, and that will also be true 3 months from now.

The situation for professionals is similar to how it is for companies. Even if AI isn't working well for you now, if you ever stop trying you will be at a big disadvantage once it does get good enough. And there is no guarantee it will ever be useful for them, so it's a tough decision to make.

No one is surprised people have moved away from the radio. But it's still important to track the progress. It has significant impact on marketing and other investments. It would be wrong to think no one is still listening to radio just because there are options many people feel are objectively better.

Did NPR not report on the recent Iran war? I have moved to podcasts a while ago but find it hard to believe NPR ignored it happened.

I had Claude Opus comb through all NPR coverage of the war over the past couple days, and it found 21 articles on their site. Its estimate was that 48% of the content was Neutral / Straight News, 43% was Negative / Critical of U.S.-Israel, and 9% was Positive / Supportive of U.S.-Israel (most articles were a combination of each).

What has your analysis of NPR's coverage of the Iran war revealed that conflicts with my narrative?

Me too, but that doesn't address the OP's concern. I use Claude Code almost daily for multiple side projects. I have the $200 per month max plan, but my ccusage metrics show I use about $1500 in API-equivalent tokens per month. And that is on top of my regular Claude web chat usage, and I run over a hundred deep research sessions per month too. I estimate that is about $500-1000 in additional API-equivalent tokens per month. API use may be profitable for Anthropic, so that doesn't mean they are losing $2000 per month on my subscription, but I bet their losses on my account are at least $1-1.5k monthly. I realize I am an extreme example, but any huge influx of users to any of these providers does mean an increase in their burn rate (not an increase to profits).

That said, none of these losses will matter in the long run. All that matters is whether they can start replacing human labor. $1 trillion in AI spending is nothing if AI can do 10% of white collar work. That would represent $2-3 trillion in value every year. The fact we have only seen about $1.2 trillion in AI investment over the past 5 years shows investors aren't that confident AI can replace that many jobs. If they were, they would be investing 5 times as much.

I am a pretty big proponent of Anthropic's models, and I lead a $80+ billion dollar company's AI platform strategy pod within our corporate strategy department. Winning over someone like me has a significant impact on Anthropic's ability to win more lucrative corporate deals. My company can afford to spend $100 per day on a continuously running agent as long as it can replace a human employee ($35k/yr vs $50-$150k/yr).

Most of it came from open source projects and other public code repositories.

There are trillions of lines of code in all public code repositories, and hundreds of millions of lines of code in stack overflow pages. That puts stack overflow at about 0.01% of the contribution to these coding models. Maybe closer to 0.1% contribution considering there is more explanation for the code in stack overflow than there is in your average codebase.

Most counties in most states do it this way, not just blue states.

We already do this in the US. Zoning is local, so it isn't consistent everywhere, but in most jurisdictions you do need to rezone agricultural land to commercial or industrial use.

Someone already mentioned that the US already has a large amount of idle and abandoned arable land, but here are the actual numbers:

Active Cropland: 338 million acres
Pasture Only: 12 million acres
Idle Cropland: 39 million acres
Abandoned Cropland: 30 million acres

Hines Research Data Center Land Needs 2025-2030: 40,000 acres

So even if all new data centers are built on arable land, we would need to activate about 0.06% of our idle or abandoned cropland. This is a non-issue.

That is 100% true. But it also isn't how trillion dollar companies get their valuations. They get it by having a deliberate product strategy and a strong moat that will defend their revenue for decades to come.

It won't be returned to families. It will be returned to the companies that paid the tarrifs and have already passed along those costs to consumers. There will be no incentive to pass down these refunds to consumers.

That is pretty sick that you are looking forward to someone getting killed.

But if it does excite you so much, I suggest tracking Amazon / UPS / FedEx delivery truck fatalities instead. You'll get to enjoy hearing about a few fatalities every month.

AI isn't propping up the US economy. It is just keeping us out of a recession. If we had a recession it wouldn't be the end of the world. It would probably be pretty healthy in the long run.

I doubt the AI bubble bursting is going to make you any happier. It will slow down the advancement of AI as much as the .com bubble slowed down the growth of the Internet and the housing bubble slowed down the growth in housing prices. The only people who'll benefit are those who didn't have much invested in the market 5 years ago and want a short period of discounted stock prices before AI really takes off. I was in my late 20's during the housing bubble so that crash definitely helped me buy a cheap house in my early 30's, so perhaps you're in a similar situation.

Have there really been that many frauds? Gemini found 7 former winners who have been charged with fraud (5 found guilty with 2 cases ongoing) out of 420 members. When you include other issues like sex offenses and workplace bullying it goes up to 15. So around 1.5 - 3.5% of their winners either went on to do bad things or were already doing them without getting caught.

Is the rest of the enterprise business world really any better than that?

I don't think many if any of the bailouts of the past 20 years were done to save jobs at the companies being bailed out. They were to maintain our banking systems so other companies could continue running, or to protect industries that would have lost market share to foreign companies.

I'm not defending the behavior. I'm just pointing out that protecting jobs hasn't been a common justification for bailouts in the past.