The market is a funnel, rewarding centralization, you can thrustbust the t-1000 all you want, but the mercury meets in the middle and guess who is back.
Yes, but that's by design in some sense. Monopoly is a carrot, entrepreneurs are donkeys on treadmills chasing after such carrots. The market's value to society in this metaphor is the energy produced by the treadmills, so while it's critical to offer donkeys the promise of a carrot, it's even more important they don't actually get it, because then they step off the treadmill and become a nuisance.
I agree you could make laws allowing for self-busting. As in, some management whistleblower having evidence of monopoly practices, reports it and the company gets autosplit- with the dissidents as manager of the new company.
Look, US tech services can either follow our laws our they can get the fuck out of our market.
The way people treat them breaking our laws as their god given right and then frame our fine as anything but their own fault is starting to really annoy me.
Respect our sovereignty, follow our laws. Or don't, but then don't be surprised that breaking the law is punished.
Maybe things are different in the US, maybe corporate capture has progressed to a point where laws are optional for megacorps. But we still have rule of law and we'd like to keep it that way.
While there is “law”, how much agency does an average European individual have in shaping those laws? This is a multi national body for which you can’t vote for its leader. You can’t vote on any of its resolutions or referendums.
It’s representative democracy, like most democracies. It is, granted, a particularly complex and confusing one.
In practice, the average European has more agency in shaping EU law than the average American does in shaping US law; the US’s electoral system means that, realistically, on a federal level, most peoples’ votes simply do not matter (two party FPTP system, with most seats being safe). The EU parliament is at least multiparty proportional representation, and most member states’ governments are also elected via some sort of PR.
> for which you can’t vote for its leader
Most democracies don’t have a directly elected head of state; it’s _largely_ a thing in the Americas and Africa. In terms of big countries with a true presidential system, you’re really talking the US, Russia, Brazil, Nigeria and Indonesia. I’m, ah, not sure that we should be rushing to emulate those paragons of democracy :)
France does have a directly elected head of state (I wouldn't want to emulate that either, though their system is not purely presidential), as does South Korea, Mexico, Turkey and many other countries really.
" it’s _largely_ a thing in the Americas and Africa"
" In terms of big countries with a true presidential system, you’re really talking the US, Russia, Brazil, Nigeria and Indonesia."
These are factual inaccuracies.
South Korea is a functioning democracy. They tend to jail their ex-presidents quite often, but then only those that do deserve it, which is more than we can say about a few western democracies which are very lenient about the criminality of politicians. Ah yes and the president is limited to serving one term.
The US doesn't have a directly elected head of state either. The demos elects the electoral college, which elects the head of state. Twice in recent memory directly against the popular vote, most recently Trump I.
For the Council, it is literally just a grouping of our democratically elected heads of state. Those we can of course vote for according to our countries own democratic systems.
For the Commission, our democratically elected national governments each send a representative, so those are as elected as any minister in government is.
For Parliament, the EU demos elects directly.
Not sure where the issue is? It all seems more democratic to me than, say, the Electoral College.
The EU just finds a new thing to fine each time. Since the scope of US consumer internet companies is so vast, they can mine €1b fines more or less limitlessly.
The argument is that the laws are so broad that the EU can always find a violation even when companies try to follow the law, and that they mostly target foreign companies. TBH I find this focus defensible: the big US companies are just so much bigger that it makes more sense to police them than insignificant local players. But it is a good cash cow.
The problem is that American companies try to follow the letter of the law, while going as much against the spirit of the law as possible. This is why they complain it's "too broad", because they can't avoid the spirit as they would like to.
Example here: Facebook changing their ToS to claim usage itself constitutes freely given, informed consent under GDPR. No it doesn't, you knew that was against the spirit, you did it anyways, you got fined.
This is fundamentally how the EU legal system works, you're expected to do what the law wants from you, not find some clever way to avoid what the law is for while ticking the right checkboxes.
Sometimes the "spirit of the law" seems to be "degrade your offerings until the European competitors who keep pestering us stop". Little consideration ever seems to be given to telling those European businesses to make a good product themselves. European competition regulators could be weighing consumer welfare more heavily.
I'm not sure I've observed any of that. American companies usually degrade out of spite in my experience - they'd rather not offer the product than to offer it lawfully.
I mean the current US president literally threatened to increase EU trade tariffs if the EU tries to get the big tech companies to pay their fair share of tax here. You can’t have it both ways.
The WTO is paralyzed since 2019. Nobody cares about it anymore:
"Since 2019, when the Donald Trump administration began blocking appointments to the body, the Appellate Body has been unable to enforce WTO rules and punish violators of WTO rules.[3][4][2] Subsequently, disregard for trade rules has increased, leading to more trade protectionist measures.[4][5] The Joe Biden administration maintained the freeze on new appointments."
This time the American tech industry is the uncolonized native America, and the Europeans are the settlers that ran out of land to take over and decided they must forcefully re-colonize Americans and our design decisions and way of life by force.
This is such a ridiculous offender/victim reversal. American tech companies are free to do whatever they want in America. In our country, they need to follow our law. If they don't want that, they can get out.
What they're trying to do is colonize Europe with their lawlessness, they're trying to deny us our sovereign right of determining our own rules.
There's no right to come to our jurisdiction and then decide to ignore our rules. Either come and play by the rules or stay out.
Yes, but the original purpose of the benchmark (simonw, please correct me if I'm wrong!) was to test whether new models were good at novel problem solving. Things they haven't been trained on. So yes, getting better at generating SVGs is great news (and it seems they have been) but this particular benchmark still strikes me as largely worthless now, unless SVGs happen to be what you care about in particular when a new model drops.
Instead, it feels like a more appropriate benchmark for the original purpose would be to come up with new, novel problems each time, and compare across all models (including previous ones).
Throwing my hat in the ring: Generate a pelican shaped crossword where all the clues are related to bicycles.
(Haiku 4.5: https://imgur.com/a/N112Nxo, I'm trying some others but it's very slow! Opus has been at it for about 20 minutes.)
I think a lot of LLM progress in the last 2 years or so is really from densifying the training data to niches like this and not so much from general problem solving capability. I guess the hope is that the deluge of super specific training data will pay off through better overall competence. cf https://www.dwarkesh.com/p/the-sample-efficiency-black-hole
I do have a private benchmark I use for that. I have a repository with an in-progress codebase and a very large planning document, I tell the new models to simply finish everything in the planning document, and I never push those changes. All of the recent models have essentially gotten it perfect which means that the models are all equivalently good for my level of needs.
I've mostly been running LLMs on my own hardware so the phrase "been at it for about 20 minutes" gave me pause. My first instinct was to ask "but on what hardware?" but I suppose one advantage of closed models is that there's a somewhat-consistent cognitive-effort-to-time ratio.
Fair, I don’t know if this helps but it was Opus 4.8 (high effort) via Claude Desktop, but through my org’s LLM gateway so hard to compare really. But it’s usually pretty quick.
How useful actually is this? It generates SVGs of pelicans on bicycles, sure, and some of them are (almost) spatially correct. But, none of them look good.
AI image generation suffers from this more generally. You can generate pictures of pelicans, sure. Newer models clearly generate images with more pelican-ness than before. But all of it is still uglier than sin. Drawing things accurately is one thing, making results that someone might actually want to use (without embarrassing themselves) is something else.
Just today I asked Claude Fable to take an SVG map and add a circle with a 150 mile radius around a specific city, cutting off the circle at the edge of certain boundaries. It came back a minute later with more or less exactly what I wanted, saving me maybe 5 to 10 minutes of photoshop time. It’s almost certainly not perfect, but I didn’t need it to be perfect, just a legible representation for a group of 20ish people.
OTOH I had Fable laying out a 3d broadcast room, and as part of the assets it sourced, it grabbed a camera. Makes sense. There were a lot of problems with the scene - backwards facing props and clipping everywhere. But the true coup de grace was a mansized, more like a giant sized, 35mm camera pointing at everything. I was seriously confused for a minute until I realized what had happened.
I used Claude to plan some landscaping, and it gave me SVG diagrams of flower beds showing where to place plants and their approximate mature size. It was pretty useful.
Even if the models don't break through any particular "uglier than sin" barrier, with a bit more work, presumably the SVGs could become importable into an editor that would let a human apply taste and discretion. Seems to me like a heck of a head-start.
As for conventional diffusion-model stuff, I happen to think there are some pieces of AI art that still look really good even knowing they're AI.
Great for charts, illustrations etc, can do much more with SVG than pretty pictures. Good spatial understanding in an SVG can be really nice for model to have
I used it to generate some icons for my app recently, I found it a pleasantly useful feature already in its current state. Before this I would’ve had to close the code editor, open something like Illustrator, use something like a tablet (or struggle with the mouse and keyboard), and export the result. For something like a generic icon for a button, or a simple geometric figure, this always felt too tedious. On the other hand, my apps looked visually impressive before the AI and people valued my artistic ability more. So there’s that… Not every software engineer also had an eye for design. With AI generation, they can bridge that gap.
In D&D, min-maxing (hyphen and one x) is already a used term for character optimization, minimizing undesirables and maximizing desirables (I guess).
Minimizing peaks is probably not a good strategy generally. Even low peaks may have some benefit ("if you know your problem is in this domain and performance is critical, this tool offers a 3% advantage").
Minimizing troughs may be more attractive. People generally seem to react more strongly to negatives and competitors can devise benchmarks which emphasize one's troughs.
While a universal expert would be convenient and broad knowledge aids some forms of creativity, specialization has substantial advantages.
Trading max performance (primary metric of concern) for some improvement in a secondary metric often makes sense and that could be an interpretation of "minmaxxing", reducing the over-emphasis of a single metric which would otherwise be maximized.
Yes, it was a subtle referencing to rpg minmaxing, and also the minimax algorithm, beyond the more surface-level idea that "maximalist" trends tend to be followed by "minimalist" ones as a natural counterreaction.
The fear is that the SVGmaxxing is limited to "X doing Y". If such 'template maxxing' exists, it will break for other templates, e.g. "X not doing Y", "X and Y doing Z", "X doing Y doing Z", etc.
Interesting! The reason I asked is I've had poor results with Fable and 3d stuff. Its spatial awareness seems poor - doing things like rotating left and then right back, and then pitching nonsensically, just to try to capture a segment of a scene for a verification pass. And its placement often results in clipping, misrotations, and so on.
It could well be that the exact domain matters more than the bigger picture concepts, like 3d. One of the ever fewer reminders that this tech is still just fundamentally a token prediction algorithm.
The failure seems to be that a scene is inherently hierarchical with a graph with offset, rotation, scale and a name - and this seems to be transferred back poorely fro example during 3d generation.
If a model can improve at drawing "X doing Y" and that prompt wasn't in the training set then it means it has improved its internal mapping from text-to-spatial-to-text.
The improvement would be limited to the template "X doing Y". It would not be a general improvement unless hundreds or thousands of diverse templates were used.
No, the point is that a general-ish ability to draw good SVGs is a useful ability in itself. People need SVGs for all sorts of purposes, and if AI can generate one for them, that's mostly useful (discussions about art and employment etc notwithstanding).
That said, I think this would correlate relatively little with general programming ability. They're not unrelated, of course, but being able to generate code that paints an accurate + esthetically pleasing image is quite different from generating code that achieves a non-spatial goal.
Yeah, if you want to generate an aesthetically pleasing SVG, you'd be better off asking a pixel-based image-generation model for "vector art" and then deconstructing it into an equivalent SVG with something like LayerPeeler. https://layerpeeler.github.io/
These are rectangles with some fill and strokes, textboxes lines and arrows. If the text has already been written, and the author has a clear idea of the flow all it takes are some a dozen or so points and click, and setting a couple of fills, then copy pastes, and finally some resizes, selects and moves.
This design is trivial. I admit it would be hard to achieve in Word (or at least for me because I don’t know how to make a diagram in Word any more) but Figma and Inkascape are made to do these things, and have optimized UI for that (personally I would have just used mermaid though).
I think you may have lost your faith in human capabilities just a little bit if you think drawing stuff like this takes any time or effort at all. Compared to designing and architecturing the system, drawing the diagram is trivial. Now I know that my parent did neither but it seems like they vibe-coded the whole thing. I’m sure they will end up with a fun little toy from the whole endeavor they can play with for 2 weeks before abandoning. Maybe the author will even feel bad about the carbon footprint of this whole exercise and buy some carbon offsets to make up for it.
I am glad you get to have fun while people you will never meet and live in Sweden or Virginia have to suffer higher electricity prices. But we all get to enjoy the warmer climate together. But at least you can enjoy your self having the plagiarizer build you a single use toy that only took the carbon footprint of entire nations to create.
AI consumed around 0.5% of the world’s electricity in 2025[1]
Where I'm from data centers help the renewable mix by subsidizing transmission from other geographic zones. I'm actually improving the environment by using it.
Agree directionally - even back in sonnet 3.5 days, I was helping some friends by showing them how to create intermediate representations for SVG building blocks mapping to parametrizable functions that can be used to make interactive SVG-rendered visualizations for various medical needs.
Huh? Open networks are open. You don't go to prison for that. Anyway, as the next comment said, there aren't many open networks anymore, though you'll still find some (see other comment)
Way less than before, but, as an example, the supermarket very close to my home (in my home country - my other comment elsewhere is about our home in Japan) has an open network. It's presumably so that when customers are there then the supermarket's app, on customers phones, will have access to coupons and the like.
My own phone utilizes that (I don't use a data plan with that phone, as I mostly use it for calls and not having a data plan saves me a ton of money - it's basically free).
I've seen a lot of open networks in stores, cafes and malls in my country too but on most occasions they at least take you to a captive portal to accept their TOS or ask you to enter an email address to spam you later, they are not just connect and browse with zero interaction.
There are still connect-and-use ones, like the one I mentioned, and the rest of them (fortunately there aren't that many which requires an email address anymore, though some exist) are just uselessly annoying. Airports, for example, always try to go through a 'connect' phase, and on Android that's sent through some Android-tool which often don't even work for some of those networks.
And for no reason - TOS is useless, and sometimes there's not even a TOS - it's just a "connect' button. All of that makes it hard, sometimes impossible to connect, for absolutely zero reason. :-(
I haven't seen that many actually. In our neighborhood in Japan the area is crowded with home networks, they're all encrypted.
And when I've tried navigating in large cities (Tokyo included) and found that my stupid tablet's GPS relied on getting the ephemeris data from the internet, and never through the GPS itself, I couldn't start navigating until I connected to something.. and I spent hours trying to find an open network, unsuccessfully. Happened twice, the second time in Nagoya. No open networks. Well, Starbucks have that, but then you need to find one.
they say that Japan is simultaneously living 20 years in the past and 20 years in the future. lots of legacy devices are still being used for all manner of quirky convenience contraptions. they also live in an extremely high trust society where networks don't necessarily need to be secured
If you want a very specific example, the original nintendo DS can only connect to WEP connections, which meant that when I was there I was able to play parts of DS games that wouldn't have been accessible in the west
This is such a petty and impotent ruling. If you want to ban them from using culture to make derivative works without proper compensation then do that.
But if you don't want to ban them, telling them to buy one book of each, likely second hand, is complete pettiness that resulted in destructive scanning of millions of books, many of which were already practically available in digital form.
>This is such a petty and impotent ruling. If you want to ban them from using culture to make derivative works without proper compensation then do that.
That's because the judges are supposed to rule on questions of law (ie. "is AI training fair use?"), not whether they think AI's good or not.
The much simpler way to avoid squatting is to make .com domains cost $200 a year. This will instantly end the vast majority of domain squatting on the .com TLD and if people can easily get the .com they need for their business then the other TLDs are not going to have much squatting activity.
> The much simpler way to avoid squatting is to make .com domains cost $200 a year
A monopolist hiking prices to this extent will likely see legal action against them. That's a 20x increase you're proposing.
It's also unlikely to have a material effect. .com used to cost $75 a year back in the day, and that didn't stop squatters, and high value domain transfer sales. $75 in 1990s dollars is about $150-$190 today.
the most fair distribution for limited sought-after resources that are inconsequential (like domain names) are raffles. let people apply in a 4-week window and then randomly assign it to one of the applicants.
then don't allow reselling, just allow giving it back and do a raffle again
First come first served is capitalism in disguise. Someone will automate a tool that watches for these and jumps on them, and the person with the most resources, can get a close network connection, etc will "buy" it.
I don't think that anyone claims that there aren't any legitimate sites on .garden, but the risk of using an abuse-prone TLD is that Bayesians are going to assign you an increased prior risk of abuse. Honestly the TLD is making more money from the abusers than from Plantura so they're not going to tighten up their ship, Plantura should probably move to a different TLD.
One theory is that Elon Musk ends up being able to build AI datacenters faster and more cost effectively than anyone else, which seems plausible given the success of Tesla Gigafactories.
Datacentres are close to being a commodity, though.
There's some stuff you can do with Combined Cooling & Power or other cogeneration schemes, and potential for cost-cutting: removing fire protection, doing without redundant power, reducing security, that sort of thing. But that's all fairly marginal in the grand scheme of things.
The more experimental stuff like running with extreme temperature gradients, extensive liquid cooling, or maybe using CO2 or He instead of air is certainly worth trying out, but unlikely to make much of a difference either.
SpaceX are claiming that they're making a big profit on their deals with Anthropic and OpenAI but that seems to assume that the cost of the compute is zero. I'm not a forensic accountant, so maybe it's okay to have written the value of that down already, but it's not a trick they'll be able to pull off more than once.
Really, the best thing you can say about them is that they've got a big pile of money so can finance capital expenditure cheaply. That gives them an edge of a few percentage points over the likes of Equinix (worth $14.2bn), but not much more than that.
Lower pay (but relatively more employees), integrated development and manufacture (they don't subcontract parts like many western and possibly japanese mfgs do), support by regional govs etc., they also don't pay licenses to licensing orgs and bit by bit all of this accumulates. Also huge competition leading to relentless cost cutting.
