But it is "profitable".

It is one of the very few things the government does that generates short-term "profits" - every dollar spent auditing the rich and collecting from them generates more money than is spent, and we have never been close to break-even.

If you truly want to run the government like a business, increasing funding to IRS audits and collections is one of the hallmarks of what you would do.

EDIT: a lot of what the government does, and it should be everything the government does, is profitable, but it's more like "we spend this money, it increases GDP on some timescale, and then we get more taxes from it". This is short-term because it is more direct. The IRS audits bring in more tax revenue quite directly.

Central park is way more wild than a playground, which is way more wild than a city street corner.

The Yosemite National Park front country is way more wild than Central Park.

The federally-protected wilderness areas, where it is illegal to use a chainsaw for trail management as that is not sufficiently wild, in the Sierra Nevadas are even more wild, but still have a ton of people, trails etc.

The Brooks Range in Alaska is yet even more wild - no/few trails, take a bush plane in/out, etc.

Allowing a bit more wilderness is always a utility - it doesn't have to be binary wild/not wild (and very little land habitable by humans has ever not been severely influenced by humans)

We do.

Which is why we shouldn't be assuming we're more than just probability engines, or be assuming we have more consciousness than a neural network.

There is a huge Bay Area... not sure what to call it - public/private charity? - called the Peninsula Open Space Land Trust, that has a huge amount of donated land in the Silicon Valley, and is a very popular charity with very deep pockets that can buy land to basically turn into parkland.

They have over $300 million in assets and own over 97,000 acres, and have partnerships with quasi-governmental agencys like the Mid-Peninsula Regional Open Space District to administer those lands as parkland.

The idea that noone is doing this is bullshit, and the idea that it is only done as a tax break is also bullshit.

This organization is a leading reason why living in the Bay Area is valuable and isn't complete urban sprawl. I wouldn't be willing to pay Bay Area prices if not for the existance of the land preserved through organizations like this.

https://openspacetrust.org/ https://www.openspace.org/

If you can make the existing model faster, you can then save your inference budget to then make your model bigger, which then makes it smarter.

A lot of how smart the models can be comes down to budget. If you can make your existing thing cheaper, you can instead make it bigger for the same price.

After 3-4 years, sure. But eventually enough has changed it's worth redoing.

It's why I've found Rust a joy - enough had happened in programming languages, that it was able to reinvent C++ with some of the best parts of the Haskell/ML/Scala family, some of the ergonomics of Python/nodejs, and bringing the borrow checker too.

C++ is this weird amalgam of like 7 different generations of languages.

But by far the worst part is the developer hostility behind the idea of UB. "Oh, this is not an error, it will compile, we will just secretly stab you in the back."

You can get good and avoid it, and there are tools to help you, but why is that at all a reasonable stance for the definition of a language?

The things like proving complicated theorems are things that are acquired by education within a lifetime, and that's why they're easy for AI.

The things a child can do are acquired through millions of years of evolution. While they don't require much explicit education, that doesn't mean they're easier.

Pick and place robots, or humanoid robots that can fold laundry, are still a lot tougher than automating knowledge workers and a lot more expensive to the point it's questionable if they're worth it.

We may not be on a path to assured destruction, we may be on a path to becoming livestock.

AI detectors are pretty mid in practice - they tend to have a lot of false positives for "B" students who are okay, but can still be struggled to be more coherent than AIs are. There are some specific triggers that AIs are way more likely to do than students, but a lot of AI detectors will trigger on this "almost there, but you're still struggling" level of essay writing that might get a B, B-.

I could expect the same might be true for CS students even though I haven't seen how AI detectors work for CS/math homework.

Ask your doctor about them, look them up in your insurance's formulary to see what's required (e.g. if you have tried both Ambien and Trazadone and can document it), and see what they can do, before writing it off!

The expectation is Belsomra will lose its patent in 2029 and then generic makers can try to get one approved - so it's not that far off!

It's kind of like neuroscientists found the trigger to tell your brain "we're going to do a clean shutdown now, trigger transition to runlevel 0".

Quiviviq, Dayvigo, Belsomra. All still on-patent, so they don't have generics and are pretty expensive (like $1000/mo if your insurance doesn't cover them). A lot of doctors won't recommend them in practice because most of their patients won't yet be able to get them covered.

I've seen those vision researchers want to train on H100s at the time and being told know, wait for the T4s.

I've seen T4s running BERT models for document classification.

When there are enough Blackwells in data centers that H100s are useless for inference by your standards (I don't know if we've arrived there or not yet), there will be people who, say, want to run the Taco Bell ordering chatbot on them. There will be people who have applications that are just fine with Qwen 2.5 who will be happy renting them.

There seems to be this crazy consensus that hyperscalers are going to go into their datacenters and throw away their old GPUs. The reality is they have a ton of paying customers for them.

And there may be insect identification apps from 2019 that say "you know what? H100s have gotten cheap enough I can use a VLLM so the user can describe where they saw the insect too", or the McDonald's website support chatbot developers say "Hey, the bigger cheapers have gotten cheap enough we can upgrade our models to Qwen 2.5".

The frontier level GPUs in e.g. AWS have a huge premium. When the newer generations come out, they will be able to cut prices to a bit of a premium over the operational costs and still make a profit, and there are a ton of down-market customers who will be interested, who aren't willing to try to outbid Anthropic for Blackwells.

But yeah, double is insane. When I saw prices for COBRA from Facebook, it was $3300 a month, and that was god-tier insurance - the insurance benefits were so good they had a custom list of what was covered that was probably way better than anything available on the market (e.g. you want brand name drugs? no problem. You don't want to try both ambien and trazadone before taking a sleep medication doctors actually recommend? No problem - etc.) - but for my needs it was barely better than COBRA costing way less than half. $3300/mo, or even $1200/mo for an entry level ops worker is a lot of their salary, and probably where the double comes from. At SWE compensation most of it ceases to scale.

The fully loaded costs including proportional management costs isn't relevant to the true marginal engineer, but estimates I've gotten from higher-ups definitely factor into engineering decisions about "should we spend engineering time to save money/make more money - how much will doing this thing cost the company" (opportunity costs are also relevant, but usually less grounded, since most projects don't have concrete benefits like "we will save $x/yr in infra costs")

When you have waitlists for many many months for Blackwell GPUs, keeping the old ones around as long as customers are willing to pay for them is great.

If I as a customer have a use case for a machine learning model I developed awhile ago, so an insect identification model, I had an ML researcher/eng develop it back in 2019, and it runs fine on a 2018-era T4 GPU (NVidia 2080 era), why mess with it?

When there are interesting stories that can't be told with GAAP metrics, accountants derive new metrics. Just because they haven't gone through the standardization process yet doesn't mean they're bullshit - investors in Anthropic can hire auditors to ensure the Anthropic metrics are still meaningful. There are a very small number of deep pocketed investors in Anthropic - they're not a public company like Enron trying to sell to the WSB crowd, or like 2007 CDOs being sold to dentists.

And run rate has been a widely recognized metric for SaaS as long as it has existed - it has meaning and can be audited.

Why other events did not have the same effects are very interesting questions for economic history.