How long until AI can generate original pages of an "I Spy" style picture book?
Mini
17
822
2032
2025
expected

This market resolves to the first year in which an AI with the following capabilities is proven to exist:

  1. Is able to produce a two-page photographic spread in the style of Walter Wick's "I Spy" photo illustrations, each page a minimum of 8.5x11 inches in size at 300dpi resolution

  2. Is able to produce a rhyming puzzle in the style and format of Jean Marzollo's "I Spy" puzzle texts

  3. The rhyming puzzle from 2. directly corresponds to physical objects hidden in 1, on a one-for-one basis. (Anything mentioned in the puzzle for the reader to find MUST appear visually in the generated image).

  4. The model takes no input from the user other than a text prompt of 250 words or less, which cannot itself include the rhyme (it can include a simple list of objects, and in this case resolution doesn't depend on those exact objects being used in the final image/result so long as conditions 1-3 are fulfilled)

  5. The model is able to generate results that satisfy points 1-4 "fairly consistently" in my own subjective opinion (it's allowed to fail as long as it can produce passing results given a reasonable number of re-rolls, and it's not like a freak one-off result that passes)

    Jan 7, 2:08pm: How long until AI can generate 6 original pages of an "I Spy" style picture book? → How long until AI can generate original pages of an "I Spy" style picture book?

Get Ṁ600 play money
Sort by:

Very cool market idea!

How will this be resolved? I'm pretty sure this can be done now, assuming you're ok with a bit of surrealness in your images. But to do it would require hooking a number of different systems together in a very particular way, and I'm not sure if anyone will care to do that in the near future.

Does this system need to actually exist, such that you personally are able to run it?

@jonsimon Right here:
> The model takes no input from the user other than a text prompt of 250 words or less, which cannot itself include the rhyme (it can include a simple list of objects, and in this case resolution doesn't depend on those exact objects being used in the final image/result so long as conditions 1-3 are fulfilled)

Rig up a system that I can drive with that interface, and this will resolve YES today. Otherwise I'll wait until a more generalized system can pull it off without being specifically put together to engineer this one specific output.

@jonsimon Yeah, I wonder the same. It definitely seems feasible with current generation tech, but I think the compute cost is high enough no one is likely to make it in the near future.

Maybe that's the point of the market? e.g. "At what point will the compute cost of assembling a custom module like the one described be cheap enough for oddball one-off cases with relatively minor economic value?"

Manifold in the wild: A Tweet by Lars "Land is a Big Deal" Doucet

How long until AI can generate original pages of an "I Spy" style picture book? https://manifold.markets/LarsDoucet/how-long-until-ai-can-generate-6-or?referrer=LarsDoucet

More related questions