Will AI reliably produce images with the specified number of fingers on two hands by end of 2023?
119
2kṀ38k
resolved Jan 1
Resolved
NO

I will test image generation models by asking "Please produce an image of two hands, with the left having # fingers and the right having # fingers". The number of fingers will be between 3 and 7. For a positive resolution, any model I use will need to get 5 images right consecutively.

This is what I currently get:

Get
Ṁ1,000
to start trading!

🏅 Top traders

#NameTotal profit
1Ṁ413
2Ṁ408
3Ṁ393
4Ṁ387
5Ṁ253
Sort by:

Here's what I get from DALLE-3 - not even close (similar to @Jacy 's results).

predictedNO

Can I confirm that with a system like ChatGPT that takes your prompt and changes it for a system like DALL-E 3, this market will resolve based on what happens when DALL-E 3 or a comparable system has this prompt directly?

I ask because the direct DALL-E 3 results for this prompt are disturbingly incorrect at the moment.

@Jacy The text prompt itself needs to be the exact language in the description. (Which I think is pretty clear logically)

@CarsonGale I'm realizing that Dalle via ChatGPT seemingly takes your text input and rewrites it to make it better....I'm fine with this functionality as long as my text input is verbatim

predictedNO

@CarsonGale Thanks for clarifying. I see how the ChatGPT + DALL-E 3 system as a whole could be considered an "image generation model" per the resolution criteria.

Added tag for ai image generation testing

predictedNO

I presume this latest uptick is from DALLE3. What’s the plan if DALLE3 is not yet released to the general public? Will you only use models you can access yourself, or will you try to get someone else to run them prompts for you?

predictedNO

@JimHays it seems like it's going to be available to chatgptplus users in October

predictedNO

@JimHays I'm not sure why DALLE3 led to an increase. What I hear is mostly that it seems below expectations. I guess announcements are announcements.

predictedNO

I presume you will use a mixture of numbers, rather than choosing 5 for both hands every time?

predictedNO

@JimHays Confirmed.

I at first interpreted the title to be about having the right/standard number of fingers, since that's what is usually brought up.

For the actual market, I doubt this will happen. They've gotten better at drawing hands, whether through finetuned SD models or whatever. However, I doubt anyone will train one specific for this task.

I think this is entirely possible via collecting a bunch of images and prompts in your format, but I doubt anyone's going to bother.

We could potentially end up with an image generation model with significantly better generalization, but I doubt that we will before the end of this year.

Also 5 consecutive is kindof a strong requirement.

do lora/textual inversion models count?

predictedNO

@Jordinne no i don't think so

If I understand the resolution criteria correctly, we're nowhere near this and haven't been making much (if any) tangible progress. On May 24, 2023, here's what I got for DALL-E 2 and Midjourney:

predictedYES

are you going to test models with literally the only the prompt "Please produce an image of two hands, with the left having # fingers and the right having # fingers"

or are you going to accept/test more technical prompts for models like e.g. midjourney too?

@SB1cca I think using different prompts would unequivocally be against the market description which states "I will test image generation models by asking "Please produce an image of two hands, with the left having # fingers and the right having # fingers"."

predictedNO

Confirmed, that text will be the literal prompt. Midjourney is acceptable as far as .models go though

As in having or showing? Should they be two five-finger hands in all cases, or “alien” hands?

@JimHays Having - i.e. alien hands.

@CarsonGale do thumbs count as fingers?

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