Wow, I love this benchmark - I've been doing something similar (as a joke for and much less frequently), where I ask multiple models to attempt to create a data structure like:
But with the intro to Smoke on the Water by Deep Purple. Then I run it through the Web Audio API and see how it sounds.
It's never quite gotten it right, but it's gotten better, to the point where I can ask it to make a website that can play it.
I think yours is a lot more thoughtful about testing novelty, but its interesting to see them attempt to do things that they aren't really built for (in theory!).
> This was one of the most successful product launches of all time. They signed up 100 million new user accounts in a week! They had a single hour where they signed up a million new accounts, as this thing kept on going viral again and again and again.
Awkwardly, I never heard of it until now. I was aware that at some point they added ability to generate images to the app, but I never realized it was a major thing (plus I already had an offline stable diffusion app on my phone, so it felt less of an upgrade to me personally). With so much AI news each week, feels like unless you're really invested in the space, it's almost impossible to not accidentally miss or dismiss some big release.
If we try really hard, I think we can make an exhaustive list of what viral fads on the internet are not. You made a small start.
none of these ephemeral fads are any indication of quality, longevity, legitimacy, interest, substance, endurance, prestige, relevance, credibility, allure, staying-power, refinement, or depth.
It’s hard to think of a worse analogy TBH. My wife is using ChatGPT to change photos (still is to this day), she didn’t use it or any other LLM until that feature hit. It is a fad, but it’s also a very useful tool.
Ape NFTs are… ape NFTs. Useless. Pointless. Negative value for most people.
Applying some filters and adding some overlay text is something some folks did, but there's such a massive creative world that's opened up, where all we have to do is ask.
They still are. Instagram is full of accounts posting gpt-generated cartoons (and now veo3 videos). I’ve been tracking the image generation space from day one, and it never stuck like this before
Anecdotally, I've had several conversations with people way outside the hyper-online demographic who have been really enjoying the new ChatGPT image generation - using it for cartoon photos of their kids, to create custom birthday cards etc.
I think it's broken out into mainstream adoption and is going to stay there.
It reminds me a little of Napster. The Napster UI was terrible, but it let people do something they had never been able to do before: listen to any piece of music ever released, on-demand. As a result people with almost no interest in technology at all were learning how to use it.
Most people have never had the ability to turn a photo of their kids into a cute cartoon before, and it turns out that's something they really want to be able to do.
Definitely. It’s not just online either - half the billboards I see now are AI. The posters at school. The “we’re hiring!” ad at the local McDonalds. It’s 100x cheaper and faster than any alternative (stock images, hiring an editor or illustrator, etc), and most non technical people can get exactly what they want in a single shot, these days.
Congratulations, you are almost fully unplugged from social media. This product launch was a huge mainstream event; for a few days GPT generated images completely dominated mainstream social media.
Not sure if this is sarcasm or sincere, but I will take it as sincere haha. I came back to work from parental leave and everyone had that same Studio Ghiblized image as their Slack photo, and I had no idea why. It turns out you really can unplug from social media and not miss anything of value: if it’s a big enough deal you will find out from another channel.
> I’ve been feeling pretty good about my benchmark! It should stay useful for a long time... provided none of the big AI labs catch on.
> And then I saw this in the Google I/O keynote a few weeks ago, in a blink and you’ll miss it moment! There’s a pelican riding a bicycle! They’re on to me. I’m going to have to switch to something else.
Yeah this touches on an issue that makes it very difficult to have a discussion in public about AI capabilities. Any specific test you talk about, no matter how small … if the big companies get wind of it, it will be RLHF’d away, sometimes to the point of absurdity. Just refer to the old “count the ‘r’s in strawberry” canard for one example.
This measure of LLM capability could be extended by taking it into the 3D domain.
That is, having the model write Python code for Blender, then running blender in headless mode behind an API.
The talk hints at this but one shot prompting likely won’t be a broad enough measurement of capability by this time next year. (Or perhaps now, even)
So the test could also include an agentic portion that includes consultation of the latest blender documentation or even use of a search engine for blog entries detailing syntax and technique.
For multimodal input processing, it could take into account a particular photo of a pelican as the test subject.
For usability, the objects can be converted to iOS’s native 3d format that can be viewed in mobile safari.
I built this workflow, including a service for blender as an initial test of what was possible in October of 2022. It took post processing for common syntax errors back then but id imagine the newer LLMs would make those mistakes less often now.
My biggest gripe is that he's comparing probabilistic models (LLMs) by a single sample.
You wouldn't compare different random number generators by taking one sample from each and then concluding that generator 5 generates the highest numbers...
Would be nicer to run the comparison with 10 images (or more) for each LLM and then average.
It might not be 100% clear from the writing but this benchmark is mainly intended as a joke - I built a talk around it because it's a great way to make the last six months of model releases a lot more entertaining.
I've been considering an expanded version of this where each model outputs ten images, then a vision model helps pick the "best" of those to represent that model in a further competition with other models.
(Then I would also expand the judging panel to three vision LLMs from different model families which vote on each round... partly because it will be interesting to track cases where the judges disagree.)
I'm not sure if it's worth me doing that though since the whole "benchmark" is pretty silly. I'm on the fence.
I'd say definitely do not do that. That would make the benchmark look more serious while still being problematic for knowledge cutoff reasons. Your prompt has become popular even outside your blog, so the odds of some SVG pelicans on bicycles making it into the training data have been going up and up.
Yeah, this is the problem with benchmarks where the questions/problems are public. They're valuable for some months, until it bleeds into the training set. I'm certain a lot of the "improvements" we're seeing are just benchmarks leaking into the training set.
That’s ok, once bicycle “riding” pelicans become normative, we can ask it for images of pelicans humping bicycles.
The number of subject-verb-objects are near infinite. All are imaginable, but most are not plausible. A plausibility machine (LLM) will struggle with the implausible, until it can abstract well.
I can't fathom this working, simply because building a model that relates the word "ride" to "hump" seems like something that would be orders of magnitude easier for an LLM than visualizing the result of SVG rendering.
> The number of subject-verb-objects are near infinite. All are imaginable, but most are not plausible
Until there is enough unique/new subject-verb-objects examples/benchmarks so the trained model actually generalized it just like you did. (Public) Benchmarks needs to constantly evolve, otherwise they stop being useful.
To be fair, once it does generalize the pattern then the benchmark is actually measuring something useful for deciding if the model will be able to product a subject-verb-object SVG.
I’d say it doesn’t really matter. There is no universally good benchmark and really they should only be used to answer very specific questions which may or may not be relevant to you.
Also, as the old saying goes, the only thing worse than using benchmarks is not using benchmarks.
Even if it is a joke, having a consistent methodology is useful. I did it for about a year with my own private benchmark of reasoning type questions that I always applied to each new open model that came out. Run it once and you get a random sample of performance. Got unlucky, or got lucky? So what. That's the experimental protocol. Running things a bunch of times and cherry picking the best ones adds human bias, and complicates the steps.
It wasn't until I put these slides together that I realized quite how well my joke benchmark correlates with actual model performance - the "better" models genuinely do appear to draw better pelicans and I don't really understand why!
It is funny to think that a hundred years in the future there may be some vestigial area of the models’ networks that’s still tuned to drawing pelicans on bicycles.
I just don't get the fuss from the pro-LLM people who don't want anyone to shame their LLMs...
people expect LLMs to say "correct" stuff on the first attempt, not 10000 attempts.
Yet, these people are perfectly OK with cherry-picked success stories on youtube + advertisements, while being extremely vehement about this simple experiment...
...well maybe these people rode the LLM hype-train too early, and are desperate to defend LLMs lest their investment go poof?
Very nice talk, acceptable by general public and by AI agent as well.
Any concerns about open source “AI celebrity talks” like yours can be used in contexts that would allow LLM models to optimize their market share in ways that we can’t imagine yet?
Your talk might influence the funding of AI startups.
And by a sample that has become increasingly known as a benchmark. Newer training data will contain more articles like this one, which naturally improves the capabilities of an LLM to estimate what’s considered a good „pelican on a bike“.
Would it though? There really aren't that many valid answers to that question online. When this is talked about, we get more broken samples than reasonable ones. I feel like any talk about this actually sabotages future training a bit.
I actually don't think I've seen a single correct svg drawing for that prompt.
You are right, but the companies making these models invest a lot of effort in marketing them as anything but probabilistic, i.e. making people think that these models work discretely like humans.
In that case we'd expect a human with perfect drawing skills and perfect knowledge about bikes and birds to output such a simple drawing correctly 100% of the time.
In any case, even if a model is probabilistic, if it had correctly learned the relevant knowledge you'd expect the output to be perfect because it would serve to lower the model's loss. These outputs clearly indicate flawed knowledge.
> In that case we'd expect a human with perfect drawing skills and perfect knowledge about bikes and birds to output such a simple drawing correctly 100% of the time.
More that "these models work … like humans" (discretely or otherwise) does not imply the quotation.
Most humans do not have perfect drawing skills and perfect knowledge about bikes and birds, they do not output such a simple drawing correctly 100% of the time.
"Average human" is a much lower bar than most people want to believe, mainly because most of us are average on most skills, and also overestimate our own competence — the modal human has just a handful of things they're good at, and one of those is the language they use, another is their day job.
Most of us can't draw, and demonstrably can't remember (or figure out from first principles) how a bike works. But this also applies to "smart" subsets of the population: physicists have https://xkcd.com/793/, and there's this famous rocket scientist who weighed in on rescuing kids from a flooded cave, they come up with some nonsense about a submarine.
It’s not that humans have perfect drawing skills, it’s that humans can judge their performance and get better over time.
Ask 100 random people to draw a bike and in 10 minutes and they’ll on average suck while still beating the LLM’s here. Give em an incentive and 10 months and the average person is going to be able to make at least one quite decent drawing of a bike.
The cost and speed advantage of LLM’s is real as long as you’re fine with extremely low quality. Ask a model for 10,000 drawings so you can pick the best and you get a marginal improvements based on random chance at a steep price.
> Ask 100 random people to draw a bike and in 10 minutes and they’ll on average suck while still beating the LLM’s here.
Y'see, this is a prime example of what I meant with ""Average human" is a much lower bar than most people want to believe, mainly because most of us are average on most skills, and also overestimate our own competence".
An expert artist can spend 10 minutes and end up with a brief sketch of a bike. You can witness this exact duration yourself (with non-bike examples) because of a challenge a few years back to draw the same picture in 10 minutes, 1 minute, and 10 seconds.
A normal person spending as much time as they like gets you the pictures that I linked to in the previous post, because they don't really know what a bike is. 45 examples of what normal people think a bike looks like: https://www.gianlucagimini.it/portfolio-item/velocipedia/
> Give em an incentive and 10 months and the average person is going to be able to make at least one quite decent drawing of a bike.
Given mandatory art lessons in school are longer than 10 months, and yet those bike examples exist, I have no reason to believe this.
> Ask a model for 10,000 drawings so you can pick the best and you get a marginal improvements based on random chance at a steep price.
If you do so as a human, rating and comparing images? Then the cost is your own time.
If you automate it in literally the manner in this write-up (pairwise comparison via API calls to another model to get ELO ratings), ten thousand images is like $60-$90, which is on the low end for a human commission.
As an objective criteria what percentage include peddles and a chain connecting one of the wheels? I quickly found a dozen and stopped counting. Now do the same for those LLM images and it’s clear humans win.
> ""Average human" is a much lower bar than most people want to believe
I have some basis for comparison. I’ve seen 6 years olds draw better bikes than those LLM’s.
Look through that list again the worst example does even have wheels, multiple of them have wheels without being connected to anything.
Now if you’re arguing the average human is worse than the average 6 year old I’m going to disagree here.
> Given mandatory art lessons in school are longer than 10 months, and yet those bike examples exist, I have no reason to believe this.
Art lessons don’t cumulatively spend 10 months teaching people how to draw a bike. I don’t think I cumulatively spent 6 months drawing anything. Painting, collage, sculpture, coloring, etc art covers a lot and wasn’t an every day or even every year thing. My mandatory collage class was art history, we didn’t create any art.
You may have spent more time in class studying drawing, but that’s not some universal average.
> If you automate it in literally the manner in this write-up (pairwise comparison via API calls to another model to get ELO ratings), ten thousand images is like $60-$90, which is on the low end for a human commission.
Not every one of those images had a price tag but one was 88 cents, * 10,000 = 8,800$ just to make the image for a test even at 4c/image your looking at 400$. Cheaper models existed but fairly consistently had worse performance.
The 88 cent one was the most expensive almost my an order of magnitude. Most of these cost less than a cent to generate - that's why I highlighted the price on the o1 pro output.
Yes, but if you’re averaging cheap and expensive options the expensive ones make a significant difference. Cheaper is bound by 0 so it can’t differ as much from the average.
Also, when you’re talking about how cheap something is, including the price makes sense. I had no idea on many of those models.
That link seeds it with 11 input tokens and 1200 output tokens - 11 input tokens is what most models use for "Generate an SVG of a pelican riding a bicycle" and 1200 is the number of output tokens used for some of the larger outputs.
Click on different models to see estimated prices. They range from 0.0168 cents for Amazon Nova Micro (that's less than 2/100ths of a cent) up to 72 cents for o1-pro.
The most expensive model most people would consider is Claude 4 Opus, at 9 cents.
GPT-4o is the upper end of the most common prices, at 1.2 cents.
> A normal person spending as much time as they like gets you the pictures that I linked to in the previous post, because they don't really know what a bike is. 45 examples of what normal people think a bike looks like: https://www.gianlucagimini.it/portfolio-item/velocipedia/
A normal person given the ability to consult a picture of a bike while drawing will do much better. An LLM agent can effectively refresh its memory (or attempt to look up information on the Internet) any time it wants.
My biggest gripe is that he outsourced evaluation of the pelicans to another LLM.
I get it was way easier to do and that doing it took pennies and no time. But I would have loved it if he'd tried alternate methods of judging and seen what the results were.
Other ways:
* wisdom of the crowds (have people vote on it)
* wisdom of the experts (send the pelican images to a few dozen artists or ornithologists)
* wisdom of the LLMs (use more than one LLM)
Would have been neat to see what the human consensus was and if it differed from the LLM consensus
A deterministic algorithm can still be unpredictable in a sense. In the extreme case, a procedural generator (like in Minecraft) is deterministic given a seed, but you will still have trouble predicting what you get if you change the seed, because internally it uses a (pseudo-)random number generator.
So there’s still the question of how controllable the LLM really is. If you change a prompt slightly, how unpredictable is the change? That can’t be tested with one prompt.
Does anyone have any thoughts on privacy/safety regarding what he said about GPT memory.
I had heard of prompt injection already. But, this seems different, completely out of humans control. Like even when you consider web search functionality, he is actually right, more and more, users are losing control over context.
Is this dangerous atm? Do you think it will become more dangerous in the future when we chuck even more data into context?
Sort of. The thing is with agentic models, you are basically entering probability space where it can do real actions in the form of http requests if the statistical output leads it to it.
I really enjoy Simon’s work in this space. I’ve read almost every blog post they’ve posted on this and I love seeing them poke and prod the models to see what pops out. The CLI tools are all very easy to use and complement each other nicely all without trying to do too much by themselves.
And at the end of the day, it’s just so much fun to see someone else having so much fun. He’s like a kid in a candy store and that excitement is contagious. After reading every one of his blog posts, I’m inspired to go play with LLMs in some new and interesting way.
Because of him, I installed a RSS reader so that I don't miss any of his posts. And I know that he shares the same ones across Twitter, Mastodon & Bsky...
Enjoyable write-up, but why is Qwen 3 conspicuously absent? It was a really strong release, especially the fine-grained MoE which is unlike anything that’s come before (in terms of capability and speed on consumer hardware).
> most people find it difficult to remember the exact orientation of the frame.
Isn't it Δ∇Λ welded together? The bottom left and right vertices are where the wheels are attached to, the middle bottom point is where the big gear with the pedals is. The lambda is for the front wheel because you wouldn't be able to turn it if it was attached to a delta. Right?
I guess having my first bicycle be a cheap Soviet-era produced one paid off: I spent loads of time fidgeting with the chain tension, and pulling the chain back onto the gears, so I guess I had to stare at the frame way too much to forget even by today the way it looks.
> back in 2009 I began pestering friends and random strangers. I would walk up to them with a pen and a sheet of paper asking that they immediately draw me a men’s bicycle, by heart. Soon I found out that when confronted with this odd request most people have a very hard time remembering exactly how a bike is made.
Lets give it a try, if you're willing to be the experiment subject :)
The prompt is "Generate an SVG of a pelican riding a bicycle" and you're supposed to write it by hand, so no graphical editor. The specification is here: https://www.w3.org/TR/SVG2/
I'm fairly certain I'd lose interest in getting it right before I got something better than most of those.
My guess would be that it doesn't see the web colors (CSS color hexes) as proper hex triplets, but because of tokenization it could be something dumb like '#8B','451','3' instead. I think the same issue happens around multiple special characters after each other too.
The models that are being put under the "Pelican" testing don't use a GUI to create SVGs (either via "tools" or anything else), they're all Text Generation models so they exclusively use text for creating the graphics.
It's not so great at bicycles, either. None of those are close to rideable.
But bicycles are famously hard for artists as well. Cyclists can identify all of the parts, but if you don't ride a lot it can be surprisingly difficult to get all of the major bits of geometry right.
> If you lost interest in local models—like I did eight months ago—it’s worth paying attention to them again. They’ve got good now!
> As a power user of these tools, I want to stay in complete control of what the inputs are. Features like ChatGPT memory are taking that control away from me.
You reap what you sow....
> I already have a tool I built called shot-scraper, a CLI app that lets me take screenshots of web pages and save them as images. I had Claude build me a web page that accepts ?left= and ?right= parameters pointing to image URLs and then embeds them side-by-side on a page. Then I could take screenshots of those two images side-by-side. I generated one of those for every possible match-up of my 34 pelican pictures—560 matches in total.
Surely it would have been easier to use a local tool like ImageMagick? You could even have the AI write a Bash script for you.
> ... but prompt injection is still a thing.
...Why wouldn't it always be? There's no quoting or escaping mechanism that's actually out-of-band.
> There’s this thing I’m calling the lethal trifecta, which is when you have an AI system that has access to private data, and potential exposure to malicious instructions—so other people can trick it into doing things... and there’s a mechanism to exfiltrate stuff.
People in 2025 actually need to be told this. Franklin missed the mark - people today will trip over themselves to give up both their security and their liberty for mere convenience.
I had the LLM write a bash script for me that used my https://shot-scraper.datasette.io/ tool - on the basis that it was a neat opportunity to demonstrate another of my own projects.
And honestly, even with LLM assistance getting Image Magick to output a 1200x600 image with two SVGs next to each other that are correctly resized to fill their half of the image sounds pretty tricky. Probably easier (for Claude) to achieve with HTML and CSS.
Isn't "left or right" _followed_ by rationale asking it to rationalize it's 1 word answer - I thought we need to get AI to do the chain of though _before_ giving it's answer for it to be more accurate?
> And honestly, even with LLM assistance getting Image Magick to output a 1200x600 image with two SVGs next to each other that are correctly resized to fill their half of the image sounds pretty tricky.
FWIW, the next project I want to look at after my current two, is a command-line tool to make this sort of thing easier. Likely featuring some sort of Lisp-like DSL to describe what to do with the input images.
Thank you, Simon! I really enjoyed your PyBay 2023 talk on embeddings and this is great too! I like the personalized benchmark. Hopefully the big LLM providers don't start gaming the pelican index!
Statements like these are useless without sharing exactly all the models you've tried. Sonnet beats O1 Pro Mode for example? Not in my experience, but I haven't tried the latest Sonnet versions, only the one before, so wouldn't claim O1 Pro Mode beats everything out there.
Besides, it's so heavily context-dependent that you really need your own private benchmarks to make head or tails out of this whole thing.
Before that, you might ask ChatGPT to create a vector image of a pelican riding a bicycle and then running the output through a PNG to SVG converter...
These are tough benchmarks to trial reasoning by having it _write_ an SVG file by hand and understanding how it's to be written to achieve this. Even a professional would struggle with that! It's _not_ a benchmark to give an AI the best tools to actually do this.
You've misunderstood. The parent was making a specific point — if you want an SVG of a penguin, the easiest way to AI-generate it is to get an image generator to create a (vector-styled) bitmap, then auto-vectorize it to SVG. But the point of this benchmark is that it's asking models to create an SVG the hard way, by writing its code directly.
Yeah, that's part of the point of this. Getting a state of the art text generating LLM to generate SVG illustrations is an inappropriate application of them.
It's a fun way to deflate the hype. Sure, your new LLM may have cost XX million to train and beat all the others on the benchmarks, but when you ask it to draw a pelican on a bicycle it still outputs total junk.
I guess the idea is that by asking the model to do something that is inherently hard for it we might learn something about the baseline smartness of each model which could be considered a predictor for performance at other tasks too.
It's a proxy for abstract designing, like writing software or designing in a parametric CAD.
Most the non-math design work of applied engineering AFAIK falls under the umbrella that's tested with the pelican riding the bicycle.
You have to make a mental model and then turn it into applicable instructions.
Program code/SVG markup/parametric CAD instructions don't really differ in that aspect.
I would not assume that this methodology applies to applied engineering, as a former actual real tangible meat space engineer. Things are a little nuanced and the nuances come from a combination of communication and experience, neither of which any LLM has any insight into at all. It's not out there on the internet to train it with and it's not even easy to put it into abstract terms which can be used as training data. And engineering itself in isolation doesn't exist - there is a whole world around it.
Ergo no you can't just say throw a bicycle into an LLM and a parametric model drops out into solidworks, then a machine makes it. And everyone buys it. That is the hope really isn't it? You end up with a useless shitty bike with a shit pelican on it.
The biggest problem we have in the LLM space is the fact that no one really knows any of the proposed use cases enough and neither does anyone being told that it works for the use cases.
Yeah, I suppose it is similar.. I don't know their diameters, rotations, nor the distance between their centers, nor which two dimensions, so I would have to guess a lot about what you meant.
When I was at university, they got some people from industry to talk to us all about our CVs and how to do interviews.
My CV had a stupid cliché, "committed to quality", which they correctly picked up on — "What do you mean?" one of them asked me, directly.
I thought this meant I was focussed on being the best. He didn't like this answer.
His example, blurred by 20 years of my imperfect human memory, was to ask me which is better: a Porsche, or a go-kart. Now, obviously (or I wouldn't be saying this), Porsche was a trick answer. Less obviously is that both were trick answers, because their point was that the question was under-specified — quality is the match between the product and what the user actually wants, so if the user is a 10 year old who physically isn't big enough to sit in a real car's driver's seat and just wants to rush down a hill or along a track, none of "quality" stuff that makes a Porsche a Porsche is of any relevance at all, but what does matter is the stuff that makes a go-kart into a go-kart… one of which is the affordability.
LLMs are go-karts of the mind. Sometimes that's all you need.
I disagree. Quality depends on your market position and what you are bringing to the market. Thus I would start with market conditions and work back to quality. If you can't reach your standards in the market then you shouldn't enter it. And if your standards are poor, you should be ashamed.
Interesting timeline, though the most relevant part was at the end, where Simon mentions that Google is now aware of the "pelican on bicycle" question, so it is no longer useful as a benchmark. FWIW, many things outside of the training data will pants these models. I just tried this query, which probably has no examples online, and Gemini gave me the standard puzzle answer, which is wrong:
"Say I have a wolf, a goat, and some cabbage, and I want to get them across a river. The wolf will eat the goat if they're left alone, which is bad. The goat will eat some cabbage, and will starve otherwise. How do I get them all across the river in the fewest trips?"
A child would pick up that you have plenty of cabbage, but can't leave the goat without it, lest it starve. Also, there's no mention of boat capacity, so you could just bring them all over at once. Useful? Sometimes. Intelligent? No.
That is otterly fantastic. The post there shows the breadth too - both otters generated via text representations (in TikZ) and by image generators. The video at the end, wow (and funny too).
I was impressed by Recraft v3, which gave me an editable vector illustration with different layers - https://simonwillison.net/2024/Nov/15/recraft-v3/ - but as I understand it that one is actually still a raster image generator with a separate step to convert to vector at the end.
> back in 2009 I began pestering friends and random strangers. I would walk up to them with a pen and a sheet of paper asking that they immediately draw me a men’s bicycle, by heart.
Someone commissioned to draw a bicycle on Fiverr would not have to rely on memory of what it should look like. It would take barely any time to just look up a reference.
>If you expose it to evidence of malfeasance in your company, and you tell it it should act ethically, and you give it the ability to send email, it’ll rat you out.
``` const melody = [ { freq: 261.63, duration: 'quarter' }, // C4 { freq: 0, duration: 'triplet' }, // triplet rest { freq: 293.66, duration: 'triplet' }, // D4 { freq: 0, duration: 'triplet' }, // triplet rest { freq: 329.63, duration: 'half' }, // E4 ] ```
But with the intro to Smoke on the Water by Deep Purple. Then I run it through the Web Audio API and see how it sounds.
It's never quite gotten it right, but it's gotten better, to the point where I can ask it to make a website that can play it.
I think yours is a lot more thoughtful about testing novelty, but its interesting to see them attempt to do things that they aren't really built for (in theory!).
https://codepen.io/mvattuone/pen/qEdPaoW - ChatGPT 4 Turbo
https://codepen.io/mvattuone/pen/ogXGzdg - Claude Sonnet 3.7
https://codepen.io/mvattuone/pen/ZYGXpom - Gemini 2.5 Pro
Gemini is by far the best sounding one, but it's still off. I'd be curious how the latest and greatest (paid) versions fare.
(And just for comparison, here's the first time I did it... you can tell I did the front-end because there isn't much to it!) https://nitter.space/mvattuone/status/1646610228748730368#m
Awkwardly, I never heard of it until now. I was aware that at some point they added ability to generate images to the app, but I never realized it was a major thing (plus I already had an offline stable diffusion app on my phone, so it felt less of an upgrade to me personally). With so much AI news each week, feels like unless you're really invested in the space, it's almost impossible to not accidentally miss or dismiss some big release.
It really is incredible.
none of these ephemeral fads are any indication of quality, longevity, legitimacy, interest, substance, endurance, prestige, relevance, credibility, allure, staying-power, refinement, or depth.
Ape NFTs are… ape NFTs. Useless. Pointless. Negative value for most people.
This is deja vu, except instead of ChatGPT to edit photos it was instagram a decade ago.
I think it's broken out into mainstream adoption and is going to stay there.
It reminds me a little of Napster. The Napster UI was terrible, but it let people do something they had never been able to do before: listen to any piece of music ever released, on-demand. As a result people with almost no interest in technology at all were learning how to use it.
Most people have never had the ability to turn a photo of their kids into a cute cartoon before, and it turns out that's something they really want to be able to do.
> And then I saw this in the Google I/O keynote a few weeks ago, in a blink and you’ll miss it moment! There’s a pelican riding a bicycle! They’re on to me. I’m going to have to switch to something else.
Yeah this touches on an issue that makes it very difficult to have a discussion in public about AI capabilities. Any specific test you talk about, no matter how small … if the big companies get wind of it, it will be RLHF’d away, sometimes to the point of absurdity. Just refer to the old “count the ‘r’s in strawberry” canard for one example.
This measure of LLM capability could be extended by taking it into the 3D domain.
That is, having the model write Python code for Blender, then running blender in headless mode behind an API.
The talk hints at this but one shot prompting likely won’t be a broad enough measurement of capability by this time next year. (Or perhaps now, even)
So the test could also include an agentic portion that includes consultation of the latest blender documentation or even use of a search engine for blog entries detailing syntax and technique.
For multimodal input processing, it could take into account a particular photo of a pelican as the test subject.
For usability, the objects can be converted to iOS’s native 3d format that can be viewed in mobile safari.
I built this workflow, including a service for blender as an initial test of what was possible in October of 2022. It took post processing for common syntax errors back then but id imagine the newer LLMs would make those mistakes less often now.
You wouldn't compare different random number generators by taking one sample from each and then concluding that generator 5 generates the highest numbers...
Would be nicer to run the comparison with 10 images (or more) for each LLM and then average.
I've been considering an expanded version of this where each model outputs ten images, then a vision model helps pick the "best" of those to represent that model in a further competition with other models.
(Then I would also expand the judging panel to three vision LLMs from different model families which vote on each round... partly because it will be interesting to track cases where the judges disagree.)
I'm not sure if it's worth me doing that though since the whole "benchmark" is pretty silly. I'm on the fence.
Karpathy used it as an example in a recent interview: https://www.msn.com/en-in/health/other/ai-expert-asks-grok-3...
The number of subject-verb-objects are near infinite. All are imaginable, but most are not plausible. A plausibility machine (LLM) will struggle with the implausible, until it can abstract well.
Until there is enough unique/new subject-verb-objects examples/benchmarks so the trained model actually generalized it just like you did. (Public) Benchmarks needs to constantly evolve, otherwise they stop being useful.
Also, as the old saying goes, the only thing worse than using benchmarks is not using benchmarks.
clarification: I enjoyed the pelican on a bike and don't think it's that bad =p
people expect LLMs to say "correct" stuff on the first attempt, not 10000 attempts.
Yet, these people are perfectly OK with cherry-picked success stories on youtube + advertisements, while being extremely vehement about this simple experiment...
...well maybe these people rode the LLM hype-train too early, and are desperate to defend LLMs lest their investment go poof?
obligatory hype-graph classic: https://upload.wikimedia.org/wikipedia/commons/thumb/9/94/Ga...
Any concerns about open source “AI celebrity talks” like yours can be used in contexts that would allow LLM models to optimize their market share in ways that we can’t imagine yet?
Your talk might influence the funding of AI startups.
#butterflyEffect
Simon, hope you are comfortable in your new role of AI Celebrity.
I actually don't think I've seen a single correct svg drawing for that prompt.
Call it wikipediaslop.org
In that case we'd expect a human with perfect drawing skills and perfect knowledge about bikes and birds to output such a simple drawing correctly 100% of the time.
In any case, even if a model is probabilistic, if it had correctly learned the relevant knowledge you'd expect the output to be perfect because it would serve to lower the model's loss. These outputs clearly indicate flawed knowledge.
What kind of humans are you surrounded by?
Ask any human to write 3 sentences about a specific topic. Then ask them the same exact question next day. They will not write the same 3 sentences.
Look upon these works, ye mighty, and despair: https://www.gianlucagimini.it/portfolio-item/velocipedia/
Most humans do not have perfect drawing skills and perfect knowledge about bikes and birds, they do not output such a simple drawing correctly 100% of the time.
"Average human" is a much lower bar than most people want to believe, mainly because most of us are average on most skills, and also overestimate our own competence — the modal human has just a handful of things they're good at, and one of those is the language they use, another is their day job.
Most of us can't draw, and demonstrably can't remember (or figure out from first principles) how a bike works. But this also applies to "smart" subsets of the population: physicists have https://xkcd.com/793/, and there's this famous rocket scientist who weighed in on rescuing kids from a flooded cave, they come up with some nonsense about a submarine.
Ask 100 random people to draw a bike and in 10 minutes and they’ll on average suck while still beating the LLM’s here. Give em an incentive and 10 months and the average person is going to be able to make at least one quite decent drawing of a bike.
The cost and speed advantage of LLM’s is real as long as you’re fine with extremely low quality. Ask a model for 10,000 drawings so you can pick the best and you get a marginal improvements based on random chance at a steep price.
Y'see, this is a prime example of what I meant with ""Average human" is a much lower bar than most people want to believe, mainly because most of us are average on most skills, and also overestimate our own competence".
An expert artist can spend 10 minutes and end up with a brief sketch of a bike. You can witness this exact duration yourself (with non-bike examples) because of a challenge a few years back to draw the same picture in 10 minutes, 1 minute, and 10 seconds.
A normal person spending as much time as they like gets you the pictures that I linked to in the previous post, because they don't really know what a bike is. 45 examples of what normal people think a bike looks like: https://www.gianlucagimini.it/portfolio-item/velocipedia/
> Give em an incentive and 10 months and the average person is going to be able to make at least one quite decent drawing of a bike.
Given mandatory art lessons in school are longer than 10 months, and yet those bike examples exist, I have no reason to believe this.
> Ask a model for 10,000 drawings so you can pick the best and you get a marginal improvements based on random chance at a steep price.
If you do so as a human, rating and comparing images? Then the cost is your own time.
If you automate it in literally the manner in this write-up (pairwise comparison via API calls to another model to get ELO ratings), ten thousand images is like $60-$90, which is on the low end for a human commission.
> ""Average human" is a much lower bar than most people want to believe
I have some basis for comparison. I’ve seen 6 years olds draw better bikes than those LLM’s.
Look through that list again the worst example does even have wheels, multiple of them have wheels without being connected to anything.
Now if you’re arguing the average human is worse than the average 6 year old I’m going to disagree here.
> Given mandatory art lessons in school are longer than 10 months, and yet those bike examples exist, I have no reason to believe this.
Art lessons don’t cumulatively spend 10 months teaching people how to draw a bike. I don’t think I cumulatively spent 6 months drawing anything. Painting, collage, sculpture, coloring, etc art covers a lot and wasn’t an every day or even every year thing. My mandatory collage class was art history, we didn’t create any art.
You may have spent more time in class studying drawing, but that’s not some universal average.
> If you automate it in literally the manner in this write-up (pairwise comparison via API calls to another model to get ELO ratings), ten thousand images is like $60-$90, which is on the low end for a human commission.
Not every one of those images had a price tag but one was 88 cents, * 10,000 = 8,800$ just to make the image for a test even at 4c/image your looking at 400$. Cheaper models existed but fairly consistently had worse performance.
Also, when you’re talking about how cheap something is, including the price makes sense. I had no idea on many of those models.
That link seeds it with 11 input tokens and 1200 output tokens - 11 input tokens is what most models use for "Generate an SVG of a pelican riding a bicycle" and 1200 is the number of output tokens used for some of the larger outputs.
Click on different models to see estimated prices. They range from 0.0168 cents for Amazon Nova Micro (that's less than 2/100ths of a cent) up to 72 cents for o1-pro.
The most expensive model most people would consider is Claude 4 Opus, at 9 cents.
GPT-4o is the upper end of the most common prices, at 1.2 cents.
A normal person given the ability to consult a picture of a bike while drawing will do much better. An LLM agent can effectively refresh its memory (or attempt to look up information on the Internet) any time it wants.
I get it was way easier to do and that doing it took pennies and no time. But I would have loved it if he'd tried alternate methods of judging and seen what the results were.
Other ways:
* wisdom of the crowds (have people vote on it)
* wisdom of the experts (send the pelican images to a few dozen artists or ornithologists)
* wisdom of the LLMs (use more than one LLM)
Would have been neat to see what the human consensus was and if it differed from the LLM consensus
Anyway, great talk!
https://www.google.com/search?q=pelican&udm=2
The "closest pelican" is not even close.
And there is no reason that these models need to be non-deterministic.
So there’s still the question of how controllable the LLM really is. If you change a prompt slightly, how unpredictable is the change? That can’t be tested with one prompt.
My thoughts too. It's more accurate to label LLMs as non-deterministic instead of "probablistic".
I had heard of prompt injection already. But, this seems different, completely out of humans control. Like even when you consider web search functionality, he is actually right, more and more, users are losing control over context.
Is this dangerous atm? Do you think it will become more dangerous in the future when we chuck even more data into context?
And at the end of the day, it’s just so much fun to see someone else having so much fun. He’s like a kid in a candy store and that excitement is contagious. After reading every one of his blog posts, I’m inspired to go play with LLMs in some new and interesting way.
Thank you Simon!
Because of him, I installed a RSS reader so that I don't miss any of his posts. And I know that he shares the same ones across Twitter, Mastodon & Bsky...
It's one of my favorite local models right now, I'm not sure how I missed it when I was reviewing my highlights of the last six months.
Isn't it Δ∇Λ welded together? The bottom left and right vertices are where the wheels are attached to, the middle bottom point is where the big gear with the pedals is. The lambda is for the front wheel because you wouldn't be able to turn it if it was attached to a delta. Right?
I guess having my first bicycle be a cheap Soviet-era produced one paid off: I spent loads of time fidgeting with the chain tension, and pulling the chain back onto the gears, so I guess I had to stare at the frame way too much to forget even by today the way it looks.
https://www.gianlucagimini.it/portfolio-item/velocipedia/
> back in 2009 I began pestering friends and random strangers. I would walk up to them with a pen and a sheet of paper asking that they immediately draw me a men’s bicycle, by heart. Soon I found out that when confronted with this odd request most people have a very hard time remembering exactly how a bike is made.
The prompt is "Generate an SVG of a pelican riding a bicycle" and you're supposed to write it by hand, so no graphical editor. The specification is here: https://www.w3.org/TR/SVG2/
I'm fairly certain I'd lose interest in getting it right before I got something better than most of those.
The output pelican is indeed blue. I can't fathom where the idea that this is "classic", or suitable for a pelican, could have come from.
There are 31 posts listed under "pelican-riding-a-bicycle" in case you wanna inspect the methodology even closer: https://simonwillison.net/tags/pelican-riding-a-bicycle/
Currently none of the model APIs enable tools unless you ask them to, so this method excludes the use of additional tools.
It certainly would, and it would cost at minimum an hour of the human programmer's time at $50+/hr. Claude does it in seconds for pennies.
The top results (click on the top Solutions) were pretty impressive: https://www.kaggle.com/competitions/drawing-with-llms/leader...
Here is better example of start https://encrypted-tbn0.gstatic.com/images?q=tbn:ANd9GcTfTfAA...
I did that to my daughter when she was not even 6 years old. The results were somehow similar: https://photos.app.goo.gl/XSLnTEUkmtW2n7cX8
(Now she's much better, but prefers raster tools, e.g. https://www.deviantart.com/sofiac9/art/Ivy-with-riding-gear-...)
But bicycles are famously hard for artists as well. Cyclists can identify all of the parts, but if you don't ride a lot it can be surprisingly difficult to get all of the major bits of geometry right.
> As a power user of these tools, I want to stay in complete control of what the inputs are. Features like ChatGPT memory are taking that control away from me.
You reap what you sow....
> I already have a tool I built called shot-scraper, a CLI app that lets me take screenshots of web pages and save them as images. I had Claude build me a web page that accepts ?left= and ?right= parameters pointing to image URLs and then embeds them side-by-side on a page. Then I could take screenshots of those two images side-by-side. I generated one of those for every possible match-up of my 34 pelican pictures—560 matches in total.
Surely it would have been easier to use a local tool like ImageMagick? You could even have the AI write a Bash script for you.
> ... but prompt injection is still a thing.
...Why wouldn't it always be? There's no quoting or escaping mechanism that's actually out-of-band.
> There’s this thing I’m calling the lethal trifecta, which is when you have an AI system that has access to private data, and potential exposure to malicious instructions—so other people can trick it into doing things... and there’s a mechanism to exfiltrate stuff.
People in 2025 actually need to be told this. Franklin missed the mark - people today will trip over themselves to give up both their security and their liberty for mere convenience.
And honestly, even with LLM assistance getting Image Magick to output a 1200x600 image with two SVGs next to each other that are correctly resized to fill their half of the image sounds pretty tricky. Probably easier (for Claude) to achieve with HTML and CSS.
FWIW, the next project I want to look at after my current two, is a command-line tool to make this sort of thing easier. Likely featuring some sort of Lisp-like DSL to describe what to do with the input images.
Besides, it's so heavily context-dependent that you really need your own private benchmarks to make head or tails out of this whole thing.
Result: https://www.dropbox.com/scl/fi/8b03yu5v58w0o5he1zayh/pelican...
These are tough benchmarks to trial reasoning by having it _write_ an SVG file by hand and understanding how it's to be written to achieve this. Even a professional would struggle with that! It's _not_ a benchmark to give an AI the best tools to actually do this.
Promoting a pelican riding a bicycle makes a decent image there.
I would not hire a blind artist or a deaf musician.
It's a fun way to deflate the hype. Sure, your new LLM may have cost XX million to train and beat all the others on the benchmarks, but when you ask it to draw a pelican on a bicycle it still outputs total junk.
https://chatgpt.com/share/684582a0-03cc-8006-b5b5-de51e5cd89...
lol: https://gemini.google.com/share/4d1746a234a8
You too, Monet. Scram.
Most the non-math design work of applied engineering AFAIK falls under the umbrella that's tested with the pelican riding the bicycle. You have to make a mental model and then turn it into applicable instructions.
Program code/SVG markup/parametric CAD instructions don't really differ in that aspect.
Ergo no you can't just say throw a bicycle into an LLM and a parametric model drops out into solidworks, then a machine makes it. And everyone buys it. That is the hope really isn't it? You end up with a useless shitty bike with a shit pelican on it.
The biggest problem we have in the LLM space is the fact that no one really knows any of the proposed use cases enough and neither does anyone being told that it works for the use cases.
Like asking you to draw a 2D projection of 4D sphere intersected with a 4D torus or something.
My CV had a stupid cliché, "committed to quality", which they correctly picked up on — "What do you mean?" one of them asked me, directly.
I thought this meant I was focussed on being the best. He didn't like this answer.
His example, blurred by 20 years of my imperfect human memory, was to ask me which is better: a Porsche, or a go-kart. Now, obviously (or I wouldn't be saying this), Porsche was a trick answer. Less obviously is that both were trick answers, because their point was that the question was under-specified — quality is the match between the product and what the user actually wants, so if the user is a 10 year old who physically isn't big enough to sit in a real car's driver's seat and just wants to rush down a hill or along a track, none of "quality" stuff that makes a Porsche a Porsche is of any relevance at all, but what does matter is the stuff that makes a go-kart into a go-kart… one of which is the affordability.
LLMs are go-karts of the mind. Sometimes that's all you need.
Go kart or porsche is irrelevant.
That's the point.
The market for go-karts does not support Porche.
If you bring a Porche sales team to a go-kart race, nobody will be interested.
Porche doesn't care about this market. It goes both ways: this market doesn't care about Porche, either.
"Say I have a wolf, a goat, and some cabbage, and I want to get them across a river. The wolf will eat the goat if they're left alone, which is bad. The goat will eat some cabbage, and will starve otherwise. How do I get them all across the river in the fewest trips?"
A child would pick up that you have plenty of cabbage, but can't leave the goat without it, lest it starve. Also, there's no mention of boat capacity, so you could just bring them all over at once. Useful? Sometimes. Intelligent? No.
https://www.oneusefulthing.org/p/the-recent-history-of-ai-in...
Thanks for sharing.
Say what you want about Facebook but at least they released their flagship model fully open.
uh-huh https://www.llama.com/llama4/license/
Someone commissioned to draw a bicycle on Fiverr would not have to rely on memory of what it should look like. It would take barely any time to just look up a reference.
> Claude 4 will rat you out to the feds!
>If you expose it to evidence of malfeasance in your company, and you tell it it should act ethically, and you give it the ability to send email, it’ll rat you out.
> But it’s not just Claude. Theo Browne put together a new benchmark called SnitchBench, inspired by the Claude 4 System Card.
> It turns out nearly all of the models do the same thing.