Will any image model be able to draw a pentagon before 2025?
294
1.1kṀ100k
resolved Mar 29
Resolved
NO

Current image models are terrible at this. (That was tested on DALL-E 2, but DALL-E 3 is no better.)

The image model must get the correct number of sides on at least 95% of tries per prompt. Other details do not have to be correct. Any reasonable prompt that the average mathematically-literate human would easily understand as straightforwardly asking it to draw a pentagon must be responded to correctly. I will exclude prompts that are specifically trying to be confusing to a neural network but a human would get. Anything like "draw a pentagon", "draw a 5-sided shape", "draw a 5-gon", etc. must be successful. Basically I want it to be clear that the AI "understands" what a pentagon looks like, similar to how I can say DALL-E understands what a chair looks like; it can correctly draw a chair in many different contexts and styles, even if it misunderstands related instructions like "draw a cow sitting in the chair".

If the input is fed through an LLM or some other system before going into the image model, this pre-processing will be avoided if I can easily do so, and otherwise it will not. If the image model is not publicly available, I must be confident that its answers are not being cherry-picked.

Pretty much neural network counts, even if it's multimodal and can output stuff other than images. A video model also counts, since video is just a bunch of images. I will ignore any special-purpose image model like one that was trained only to generate simple polygons. It must draw the image itself, not find it online or write code to generate it. File formats that are effectively code, like an SVG don't count either; it has to be "drawing the pixels" itself.

Get
Ṁ1,000
to start trading!

🏅 Top traders

#NameTotal profit
1Ṁ11,191
2Ṁ8,371
3Ṁ3,480
4Ṁ2,537
5Ṁ1,398
Sort by:

Was this market resolved correctly??? I am salty

@Guilhermesampaiodeoliveir There wasn't any proof from the YES side.

@bohaska I bet wrong, idk if i believed this or i misbetted.

What? Haven't It been already 2025 for a while? And did it happen before 2025? Why is it being resolved now?

@Guilhermesampaiodeoliveir I was late to resolve this, sorry. I'm not sure what the correctness concern would be though, it resolved NO...

@IsaacKing oh god i bet YES, mfw I did the mistake.

A more challenging market for this year:

It’s a few months late but 4o’s image generation got it right for me first try.

@bence plus an alternative description

Oh hey, ChatGPT can do octagons too!

@IsaacKing yeah, shame it didn't make it on time in 2024. It's not a diffusion model, but the market didn't specify the strategy it has to use for image generation.

Grok 3 actually can consistently draw a pentagon, but only if I use the word "pentagon". It fails as soon as described differently, like calling it a 5-sided shape. That's by far better than DALL-E or any other image model I've seen, which can't even do it with the word "pentagon".

(Grok 3 only came out in February of this year, so even if the alternative description requirement weren't there this still would have resolved NO.)

As far as I'm aware this should resolve to NO, though it looks like some models are getting close. Does anyone want to propose another image model that I should test before resolving this?

@IsaacKing I don't think it's good enough, but https://www.recraft.ai is the closest I've found. I tried "pentagon" with different source image material and got pentagonal results about 60% of the time. Mostly home plate shaped (three right angles, two 135 degree angles), a few regular.

@IsaacKing Time to resolve, any new results at this point aren't "before 2025"

ChatGPT-4o appears to do this, but in fact is writing and evaluating python code to do it. Which is IMO an extremely good solution to the problem, but does not meet the terms of this question.

sold Ṁ3 YES

Doesn't really work without the word "red" though

@AndrewMcKnight So if you include other colors, it fails?

@IsaacKing I only tried with and without the word "red". Not other colors

Haven't tested this thoroughly, but I think new Aurora model in Grok knows the 2d shape.

@TimofeyValov Promising! Assuming these are representative, it looks like it knows the word "pentagon", but can't handle a description of a 5-sided shape, which is not sufficient to resolve this to YES. Getting there though!

@IsaacKing seems like aurora does not understand the '5-gon' and really wants it to be 3d

edit: oops, lots of people already tried this below.

Did anyone try pentagonal cage fight? Very hard to get it to do anything but 8 sides, but sometimes I can get something slightly different

asking for a square cage gave me either 4 or 6 sides, it's hard to say:

And I'm not sure if this is a 2 on 1 fight or a free for all.

© Manifold Markets, Inc.Terms + Mana-only TermsPrivacyRules