radu_floricica

What this does is make it harder for new AI companies to train from scratch.

Anthropic should take it from its marketing/strategic budget. It just bought itself a 1.5B moat at exactly the time it can afford it.

To prevent misuse, businesses relying on these exemptions must provide proof (e.g. documents or test results) and publish annual reports on what they have discarded.

I wonder if anybody is keeping track of everything a mid size business needs to take care of. Each particular report probably sounds like a reasonable request, but by now they're probably well into hundreds, and they're all outside the actual scope of the business (e.g. it may seem manageable for the bureaucrats designing them, because that's what they deal with all day, but not for a small organization doing... something else).

Is supression of opinions and emotions even a good idea?

This is the wrong question. Having free opinions and emotions is such a basic and fundamental good, that any restriction needs to be a lot more than "a good idea" - it needs to clear a lot of thresholds. So it's more like "is it such an undoubtedly good idea with such a large margin that we want to take the risk?"

Ideally, one would have a meta cognitive toolkit and regularly retest their beliefs, sincerely. But that's unfortunately advanced technique, and likely requires the better part of a decade of reading and practicing.

What helps most is probably having more antibodies "in the water supply". If we spread concepts like this widely, they'll help once they're common ground for enough people. And I think it helps a lot that PG didn't give examples in his essay. Giving examples would immediately get him coded as x-ist, and color the article as political. And politics is the mind killer...

What you're doing now is taking a reasonable standard when going 200 on the autobahn and applying it everywhere else. And you don't see any problem with this because... think of the kids?

That's not how you optimize an important metric. At all. You look a how exactly those 60 people die and make targeted interventions with numbers backing them up. Otherwise best case you're doing security theatre, and most likely fuck things up with second order effects. Like things beeping all the time and distracting people - I already can't tell why my car is beeping at me.

Considering how skewed tax participation is, this would be a very one sided view. Just tax the top 49% more, no matter what's their current level of taxation, and redistribute to the lower 51%. It'll always make this criteria look like a success.

Problem is, this creates systemic effects. If you look longer term, a society that does this will end up a lot poorer than one that doesn't. Even for the bottom 51% you were optimizing. Because there are two variables to control: the redistribution, and the actual productivity. If you just focus on splitting wealth, you stop growing wealth.

Growing wealth on the other hand will make everybody richer, including the botton 51%. Simply participating in a richer economy has advantages. Plus the smaller redistribution percentage will actually end up bigger in absolute terms.

Could just be infra. I'm betting Anthropic is much better prepared.

On the other hand, when I visit Venice - which is as tight a city as can be, small streets with stone in every direction except the sky - they somehow manage to drop trees in stone squares.

Same shock, different direction, much nicer.

... carbon is fungible. Are they sure the effort of recycling them isn't higher than the effort of producing a single modern GPU?

Well, I'm pretty sure ambulances occasionally run people over. Doesn't mean we should ban them.

Orders of magnitude matter.

My biggest problem with Fable is that it includes health into its biology restrictions. Which means half the use I'd get from it ... doesn't exist.

I'm not as bitter as I could be. I'm actually quite surprised at the sanity of not avoiding the health topic completely - I think only OpenAI had a few months where ChatGPT was tip toeing in any health related conversation. Otherwise it's been almost completely ungated, and it saved and helped countless lives.

I really wish they'd find a way to ungate health and legitimate research topics.

For consumer products, definitely. Any unreasonably large charge (with a comfortable margin, like 10x) should be waived by default, unless the client specifically requested the limit be removed.

A steelman of this would be that they left OpenAI to build a company more focused on safety, and they're doing exactly that.

Well, yeah, which is why it's evil. Socialism I mean. How else would you call failing to do basic utility math while insisting you should govern and shape society?

My answer to the trolley problem is that you're allowed to not kill... unless you're the railway manager. If you're in a position of authority you pull the shit out of that switch, and then drink yourself to sleep at night. This is what authority means, not choosing the "feel good, ignore the people that could have been saved" path.

Those who are excited are a lot less likely to be out of a job in a year.

I'm not saying that's _why_ they're excited. But it's a great time to be a builder, and a terrible time to be a worker ant.

WTF is Africa doing?

Exactly what Ukraine was doing before the war, only more of it. A mixture of low level corruption, kleptocracy and outright feudalism.

Buy "Cursor", not "Cursor's IP". This means brand, users, and a shitton of data.

And if you combine a shitton of data with a lot of compute, large userbase and good engineers, you have a pretty good chance of doing something interesting.

This only means the frontier is growing faster than the price is decreasing. It's just the sum of two separate tendencies, and has little predictive value. TBH, I'm ok with this tradeoff - higher capability at slightly higher cost is perfectly fine.

2, 3 and 4 apply to the hantavirus as well.

Max x20 usage is so cheap, that it's pretty obviously subsidized. And the non-interactive usage is the easiest to explode. They could play games with "reasonable use" and whack a mole accounts that are obviously farming it, but their approach is ultimately more fair.

And I say this as somebody who just discovered agent orchestration and would absolutely love their limits to remain as they are.

I still don't get how this works. My world image must be pretty off at this point if this kind of thing is possible. A tanker is big, expensive, and not exactly easy to misplace. And for a nation to be able to send this kind of expeditions it must be both dysfunctional enough to allow this, but competent enough to be able to mount it. And other countries allow it? Why? Again with the "expensive and hard to misplace".

The advantage of a laptop is exactly that you can easily host it at home, and own everything. I have one - with an UPS also holding the router and fiber optic and an external HDD. I'm actually working right now to version 2.0 which is a beefed up version - still used laptop (found a great deal on a lenovo P1), but slightly more expensive and I'm waiting on some parts to upgrade. Should be able to even hold the production environment in a pinch.

Ah, and obviously you put a claude/codex on it, so your actual work is just ... installing claude, and maybe a linux. The rest is done by the AI - setup, scripting etc.

As a colocated option... I see it work for some people. But it'd be a niche offering, when the whole value proposition is "make my own, with blackjack and hookers".

It's 100b just to begin - the full bill would be multiples of that.

And there are options now.

This is kindof the opposite? Man + AI > either man or AI. I'd say "learn to work with Claude" is the better lesson here.

“Our technology was created to support movement, not conflict,” Hypershell told Popular Science. “Hypershell exoskeletons are designed for civilian use, from outdoor recreation to mobility support and professional applications such as search and rescue. Hypershell does not market or sell our products for military use and we do not support or condone any military application of our technology.”

That's a really stupid way to reject free advertisement. If UA picked them of all exoskeleton providers, that's decent signal. If it passes the pilot program and gets into regular use, even in a niche, that's strong signal. Either way, I'd call them to offer support, not hint that they'll refuse to sell.

Poe's law? Can't really tell, with the internet these days. Things are so polarized that people talk from and to their tribe and the message is often understood to be obvious.

Anyways, in case you're serious: there is a famous thought experiment about healthcare: should a hospital administrator approve a complex and expensive treatment to save a 7 year old girl, with a 100% success probability? Or more to the point: is approving this a "good" act, or a "bad" act? The unintuitive answer is that it depends on the opportunity cost long term, and the math is far from obvious. The quick answer is often "not even wrong" - it simply ignores a lot of facts down the line. The same cash can be used very boringly to do maintenance or to buy a piece of life saving equipment.

And it gets very dirty very fast. The easy version is pay for the operation vs buy an MRI machine - in this case you can at least compare apples to apples, if you squint - an MRI machine also saves lives. But if the alternative is renovating a waiting room, you're really off the deep end. Because doing it one time it's an obvious decision: just save the life. And it's not even bad as a general policy: have waiting rooms be a bit dusty, if this means spending more money on treatments. But... how much to cut, exactly? And then you have second order effects: throwing a moderate amount to waiting rooms can make them look a lot better, while having it linger in disrepair can make people actually avoid the hospital (if they can't even paint the walls, why would you trust them with your life?).

And all of this is assuming the hospital admin actually has mental bandwidth and latitude to make this decision. In practice, he's looking at the kid's mom when he has to turn her down because the operation money would pay for renovating a whole hospital wing. And the mom is an influencer with a large following.

democratize the means of production

This means having humans make such decisions, and even worse, it means a committee or a mob will make such decisions. Zero skin in the game, all vibes and feelings. Some girls will get saved, but I've seen how hospitals look in such a system, and it's not good.

To note: the US system may or may not be "democratized", but it managed to have exactly the same flaws - most decisions are taken by humans in a bureaucracy. The government simply wrote the rules then outsourced the bureaucracy and the blame to private corporations.

Yep. And they should also raise their expectations and work on delivering better vibecoded apps. Likely also with automation.

People are very pessimistic here in the comments, but I see no fundamental, long term reason why AI generated code can't be refactored, maintained and tested by AI just as well (or better) than average-quality human generated code. Especially because things are evolving - by the time the projects will need to be maintained, there will likely already be better tools to do that. So while I wouldn't vibecode drivers for life support systems yet, there is significant runway of tech debt for most use cases.

I don't like using these terms without qualifications, just like socialism means three different things in three different contexts.

But saying the Democratic party, with AOC, Bernie Sanders and two decades of progressism is "right"... you might as well say the sky is green. That's just ignoring any meaning of the words, not trying to find a more precise one.

Which is why you don't do it nilly willy. There are plenty of hard decision in the real world, with real consequences when you guess wrong.

But as far as my personal opinion goes, I'd prefer a bit more intervention in the world. We actually created United Nations with this purpose, but it got hobbled by Russia and China's security vetos, and by the arab block making it a "resolution against Israel" machine.

But we never decided as a planet to just leave Sudan-like atrocities to happen without taking action because "sovereignty". That's not a thing that happened, and I'm actually a bit puzzled everybody acts like we did.