Maybe that is the mojo, spending 250M and making something ugly. It’s like a rich person buying a banana stuck to the wall as art for $1M. They are showing the world they can do it not that they think the banana is delicious or pretty.
Yes indeed. Not only that, Apple worked with Android to make it work there too. It makes it a lot easier on the thief so they don’t have buy/steal an iPhone, too.
"but whatabout" is cute but how does it relate to a Singapore container ship with 20 crew members being shot at? It's like you're expecting an attack on that ship to undo or counter-balance the attack on the school? I don't get it.
Any ship choosing to pass through the Strait of Hormuz must understand that it passes through a war zone and sinking is a possible outcome.
That was their choice.
At the beginning of the war, USA has boasted about sinking a large number of Iranian ships, including unarmed ships with a much bigger crew (e.g. 104 crew members were killed when the unarmed frigate Dena was torpedoed without warning near Sri Lanka while returning from India; USA not only did not attempt to rescue the survivors, but later it even attempted to prevent the repatriation of the dead bodies).
There is no doubt that Iran has sunk far less ships than USA in this war and with many times less human casualties.
After USA earlier this year, both in the Iran war and in the naval blockade of Cuba, has amply demonstrated that they do not care about any international laws or treaties, no American has the right to criticize Iran or any other country for illegal actions, because they do exactly what USA has taught them to do.
When you do something bad and someone imitates you, you cannot condemn that action and claim that it is whataboutism when someone else reminds you that you did and continue to do the same. Whataboutism would have been only if a third unrelated party had done something similarly bad and it would have been presented as an excuse, while you, the critic, were innocent and not the original law breaker.
With its Iran war, USA has not attacked only Iran, but it has attacked the majority of the countries of the World, where this has forced price increases in a large number of products, stealing money from the pockets of billions of citizens, which was a very predictable and obvious consequence of the US actions.
Not sure about this ship, but US Navy shot plenty of civilian ships, those associated with Iran. At this point US isn't even trying to pretend their actions are lawful, by going to UN and trying to pass some resolution against Iran, they are just "we do not like Iran, so we'll strike their ships". But when Iran does the same MAGAs suddenly remember that international law exists.
But speaking about this ship the GP post implied that US it's American's fault. I mean, it may be, we don't know who shot it.
> But when Iran does the same MAGAs suddenly remember that international law exists.
I looked again and the article doesn't quote any MAGAs. I doubt US will go and complain to UN about it? It will probably just shrug, let Singapore defends its ships and sailors.
That's why I stay away from trying Gemini or doing anything much with Google until I still host my main gmail account there. One misstep and I'll find myself locked out. Because it's Google's job to spy on everyone, if I have other accounts there, those will also locked out too ("oh! you're trying to circumvent our ban? we noticed!" kind of a thing). Then the only recourse from what I understand is to make an HN post and hope it makes it to the front page and someone from Google who knows someone else at Google will help you.
Fun fact: Python natively supports complex numbers
In [1]: 2+3j
Out[1]: (2+3j)
In [2]: x=2+3j
In [3]: y=1-5j
In [4]: x+y
Out[4]: (3-2j)
In [5]: x*7
Out[5]: (14+21j)
In [6]: x*y
Out[6]: (17-7j)
In [7]: x**2
Out[7]: (-5+12j)
I've used them before for 2D graphics as a "hack".
NGL, I think this might be my favorite benchmark, ha. While the Chinese labs asses are a bit off, I certainly have big hopes for their next versions. Their hearts are certainly in the right place.
Beyond SVG, these things have been RL'd on 3d reconstruction (which carries over), and I expect given the public reaction to recent frontier advancements in 3d capabilities that this will get allocated a larger share of the post training budget in the future.
What I'd kill for would be an app where you take a pic of scribbles and drawings on a whiteboard/notebook and the app turns it into SVGs you can then continue to manipulate digitally on PC/tablet.
They can represent something in vector, but it's always very hard to use the result for anything else. Like, the paths doesn't make sense, aren't grouped to the logic elements etc.
Last time I tried, it did not. That's why there's a place on the market for an app just tuned to handle the transition seamlessly, rather than having you fucking around with prompts in LLMs.
I for one don't want a whole app for that, not anymore. It's a small task whose result is an input to something else - ideal for a tool for LLM (or script) to use.
Everyone says that until ms, apple, google, etc introduces this polished feature as part of their tablet imaging editing/document collaboration office suite. Then everyone can;t get enough of it.
> It's a small task whose result is an input to something else - ideal for a tool for LLM (or script)
Have you actually tried this? If it's a small task then why can't LLMs actually do it? Kind of like the pelican test. Every LLM can do a SVG pelican but most of the results are unusable in practice for production/productivity purposes.
Every LLm can extract some half-broken SVG hallucination of your jots, but I'm asking for an app that's polished and works accurately 99% of the time, which means a dedicated app/model tuned specifically for this task, and nobody makes this yet.
> Everyone says that until ms, apple, google, etc introduces this polished feature as part of their tablet imaging editing/document collaboration office suite. Then everyone can;t get enough of it.
I don't feel like I'm a representative sample of general userbase, so I won't dispute this too hard.
For me, it's rare but it does happen. Still, 90% of my mobile experience is "how is this so broken, why is it missing the obvious follow-up feature that completes a workflow, and why, again, am I not using a laptop for this?".
> If it's a small task then why can't LLMs actually do it?
What if they actually can?
This is exactly what Karpathy was talking about the other day[0]. If your only LLM experience is the default chat app in chat mode, then this doesn't make sense. If your experience includes using agentic harness like Claude Code/Cowork / Codex/ChatGPT Work, then it's obvious.
The LLM app in "chat mode" will just zero-shot you a pelican, or digitize your notes into SVG, and maybe improve on it a bit if you complain about the result. The same app[1] in "agentic mode" will digitize your notes, refine the SVG against the original image, reduce it to probable strokes based on high-frequency patterns in the photo and knowledge of how people write, OCR it, save both the text and SVG to your library, then extract parts you wanted and insert them into a document, typeset the rest of it, and print it on your printer - all in response to a single prompt.
LLMs in agentic mode are very good at stringing long and advanced workflows; today, if you give them shell access or computer use, they will use regular apps for you if that's the easiest way to accomplish the goal.
> Have you actually tried this?
Yes![2].
That's why I can confidently say: there is no limit to this pattern. LLMs can handle apps just fine; giving them tools just saves time and tokens.
Another example: I'm currently resurrecting a 30 year old game, and the other day I asked Claude to hack on a way to run it headlessly, so I can use my computer while it plays with it. It did, and then immediately leveraged it to just causally run several instances of the game in parallel, entirely in the background, whenever it needed to playtest something.
[1] - Claude and ChatGPT now have both modes in their mobile, web and desktop apps. I personally use Claude Code a lot, but exclusively through these surfaces (I can't be arsed to deal with TUI at home anymore). IMO, these tools are already within reach of regular users - the users just didn't notice it yet.
[2] - Not with SVGs, but equivalent with voice recordings, and yes, Claude Code set up and is managing the whole local processing pipeline. Recordings go in, Markdown (or more advanced signal processing) comes out, and in itself, it's not a task, but merely a step.
That's a very long-winded way of saying no, you haven't tried it. The point of converting something to svg is to be able to manipulate the elements, not just having an svg that mimics pixel perfect something. That's been possible for ages before llms. But the result is hardly usable if it's made up of random lines and arcs, not how someone would actually have drawn it.
Have you seen a typical SVG? No one except maybe LLMs make those semantically. It's usually just whatever InDesign or Inkscape vomited out, given construction that's about as semantic as formatting in a typical Word document.
Can you please slow down and read what's being written? Didn't even mention semantics.
One fictional example of what I mean: lets say I draw someone holding a purse. That would be a shape of their torso, with the purse overlaid. I could move those independently, and moving the purse would show the torso underneath.
However, using a png to svg, it would often just outline everything. So it's not a torso shape with a purse in front. It's a purse with a torso cut out around it, maybe even merged together as lots of individual shapes. If I were to delete the shapes making up the purse, it could be quite random what was left. Maybe some of the shapes did a double duty. Maybe it's a thousand small squares to make the purse. Maybe there is just a hole behind the purse etc.
So in the latter case, the svg isn't made up of logical elements, it's just made up of shape after shape after shape to mimic the pixels. But it's not usable as vector.
> How many pelican riding bicycle SVGs were there before this test existed? What if the training data is being polluted with all these wonky results...
But Simon has demonstrated before that asking for any animal doing something silly will still get similar results from the same model.
Do you think this might be a kind of subconscious origin of your idea? A second of a move you saw in your early life / teens stayed with you as an example of an absurd thing to draw?
It's possible, but I think it's more likely that I was inspired by the TikZ unicorn thing, my love of pelicans (we have thousands of them in the harbor ten minutes from my house), and my knowledge that bicycles are difficult for humans to draw.
It's still clearly not a part of any model's training set in the sense of pelicanmaxxing in particular. It's either a capability to generate arbitrary SVGs (or even something broader), or no specific training at all.
Certainly not because all labs are pelicanmaxxing on the same exact image. Most patterns are similar everywhere. That's the result of training on the same datasets and scraped web contaminated with other models' outputs, combined with similar training methods and resulting mode collapse. Try if e.g. dolphins knitting socks in spacesuits are similar on all/most models.
> It's either a capability to generate arbitrary SVGs (or even something broader), or no specific training at all.
> That's the result of training on the same datasets and scraped web contaminated with other models' outputs
Maybe I'm misinterpreting though, or just semantics. I would consider contamination, or whatever the first example was that they all matched, as specific training data, along with all the duplications of. I doubt it's model collapse to the same pelican, without the same original example.
Perhaps you're right, this would probably take some rigorous forensics to resolve definitely. Somehow all models have very similar patterns in text, seemingly at random. I don't think everyone is trying to train on specific patterns on purpose, it's a pretty general phenomenon. (I'm talking about mode collapse, not model)
Certainly true but I think the question is “does targeting the SVG pelican mean that the quality of all SVG output improves?” If so it’s a worthwhile thing to target.
It's a little bit like any benchmark, especially microbenchmarks. It is useful as far as it reflects the broader use cases, but the risk of overfitting to the specific task and not the broader ones is very high.
Sadly, this is all that managers know to do: take a metric and try to optimize the heck out of it. They don't care about research, technology, "don't be evil", etc. And Google is run by managers now.
Expect Google models to be benchmaxxed as much as possible, but don't expect commensurate performance on real-world tasks.
While this might be overly cynical, it's also a powerful validation of the exact feedback loop that you want to have in an organization.
If what you say is true, then the next set of benchmarks will be written to illuminate the true reflection of real-world tasks. Then those get benchmaxxed so managers can show "numbers go up."
Repeat enough times, and you get materially and genuinely improved products.
The burden is on the press, outside researchers, and general tech industry to both write and care about the metrics that matter.
There's just one problem - culture eats strategy for breakfast. You cannot govern culture or improve culture with metrics.
In large companies (Fortune 1000 etc.), you don't have much of a choice, the culture was eroded long ago, and corporate governance by metrics is a lesser of evils. But thankfully, not all companies are there yet.
reply