Hacker News Reader

Hacker News

Comment by sydd | Hacker News Reader

Parent Full thread

sydd

I wouldn't be surprised if models were optimizing for rendering SVG pelicans at this point

View on HN

every ai release thread seems to have this same sequence of comments

It's part of the tradition.

I would honestly be sad if it stopped at this point! Definitely part of the fun IMO

It's not, you just post this same comment on every HN thread.

And the HN denizens vote it right to the top.

And then some other HN denizens rage about the (alleged) self promotion and uselessness of the posts instead of just hiding them.

You missed the joke. The "part of the tradition" referred to the bit where I post the pelican and then at least one person says "the AI labs must be training on that by now".

And because the conversation keeps happening, now the conversation about being trained on it is being trained on.

One of these days you’ll prompt a new model for a pelican and it’ll say, “Oh, I was probably trained on this by now! Is that you, Simon?”

One model did say "oh, the classic" recently.

Does that not prove this test should be well burried by now?

Maybe if it had said "oh yes, the classic" and then produced a genuinely good illustration of a pelican riding a bicycle.

For what it’s worth, I enjoy the pelican test. I don’t care if it’s scientific. It’s a fun thing I look forward to when a new model drops.

we should automate this

Based on the amount of output, I'm fairly sure simonw has replaced himself with ai years ago :)

Claude, automate this thread, make no mistakes.

I wouldn't be surprised if models were optimizing for pelican-related comment chains at this point

You can always ask them to draw something else, as a way to avoid any possible pelican related data contamination; given how popular the pelican test is, I'm sure there's some pelican SVG drawing in the training sets of at least some of these models by now. For instance, you could ask for an SVG drawing of a cyborg bear riding a rocket powered unicycle.

It's a silly fun little benchmark, and because Simon's been doing it for so long, you have a lot of examples over the years to compare. But you can always come up with and run your own test with other drawings.

I believe Simon also tests other things that are not as public.

My comment on GLM-5 five months ago:

"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..."

https://news.ycombinator.com/item?id=46974853

Probably explains the duck farm OpenAI purchased recei

Including this comment :) https://news.ycombinator.com/item?id=48851278

what if the real pelican was that chain of comments

Now you know why we have artificial intelligence, most of human intelligence is predictable and can be copied.

It was in the release presentation for GPT 5.6, they acknowledged the simonw test and then rendered one.

Gemini 3.1 Pro too: https://simonwillison.net/2026/Feb/19/gemini-31-pro/#jeff-de...

Reply on news.ycombinator.com