🏅 Top traders

Will resolve NO in a few days. If you think that's wrong post a link to the model here.

wait disregard

For the people posting links to single videos: please note the description specifies ads plural. For an AI system to resolve this question YES it needs to have some general ad making ability. It does not need to be able to generate every single type of ad imaginable, but it should be able to generate a wide variety of ads (different styles, different products and product categories, targeting different demographics, etc). Nothing public comes close to this afaik. Personally I think we are 3-6 months from having decent prompt -> 2 minute video.
I am no longer going to respond to links to single videos asking if this counts. Link me to the AI or nothing.

@vluzko Does a music video that is showing the ability of someones use of AI generation of video animation & song count as a "ad"

How about this?

https://x.com/the_adprofessor/status/1732284975599271964?s=46&t=5v7DXyS3NTmwdigi0CVJfw

Oops nvm somebody beats me to it :)

@firstuserhere oh crap, this is start of 2024 not end of 2024. that explains my confusion now..

(tbc still think say 5% is too low, but betting it up to 25% was dubious)

@TheBayesian

chatgpt vindicates me (jk i should read market details before betting LOL)

@TheBayesian Yeah, I can't think of many markets where I'd believe below 10% with this much market ambiguity at play.

And the "by [year]" thing was the result of a debate a couple months ago, I think, where there was a market that seemed to be 'tricking' a lot of people with that phrasing.

@Jacy ah interesting. on the discord there were a few discussions of "nudges" to provide to users about clarifying stuff that are common ambiguities during market creation; this could be a use case, but to be clear I don't blame the market creator at all, this is 100% on me (and ig im still profitable on the market overall so doesn't matter)

https://x.com/rex_woodbury/status/1731810859809157360?s=46&t=B66Otgh2q0Cl91N3A5P9ZA

Frankly if that doesn’t count I would have a hard time imagining what would.

That sequence has a human-made soundtrack and a huge amount of editing (not just in putting brief clips back-to-back but in the clips themselves). It's implausible that it was made through a single prompt to a model, and the criteria says "good at generating complex video with sound from relatively simple prompts" and "specifically prompting it to generate ads," and @VincentLuczkow commented on a similar case below, "it remains to be seen how cherry picked those are and whether they can do more than 3 seconds."

@firstuserhere I was going to bet immediately after posting this comment, but you beat me by 2 seconds :)

@Jacy Let's look for some points of agreement:
1. If this video was made entirely by AI as the output of a prompt however elaborate or after however many tries without intermediate human assistance/editing it would meet the quality bar.
2. It being composed of clips some of which are jumpcut together would only matter in terms of the above requiring human intervention to edit it, since there are many human-made ads that likewise jumpcut.

Would you agree with those points as stated?

@alexlitz Thanks for helping unpack this.

1) I think it'd be a close call. First, there are obviously many things about the freeze frames that differ from human-produced images, such as "Adidas" not being clearly written or spelled correctly. Second, while many ads are relatively simple (e.g., no plot), there aren't many such ads that I know of that have this magnitude of visual clutter; similarly, the shoes and clothing vary a lot, which seems uncommon, but I'd happily defer to an advertising expert here or even just people who watch a lot of commercials/ads since I watch very few.

2) I don't know exactly what you mean. (i) I think a "reasonably good video ad" could have this many jumpcuts, though it's surely not that common, but the criteria suggest it needs to be the generation of multiple (many? diverse?) ads. (ii) I think systematically producing and assembling clips like these is a harder technical problem than producing reasonably good individual clips.

I also need to get to work so won't be on Manifold much for a while.

@Jacy For 1, a lot of those seem to be objections that the type or style of the ad is different than most other ads which doesn't seem to me to directly pertain to the resolution criteria, it seems clear to me that the first ads to meet the bar of being reasonably good will do that by being strong in some areas week in others and be composed in a way that leans into where they are strong. Namely, here they are strong in having compelling visuals and weaker in long continuous cuts, and thematic consistency. It also seems like considerations like the magnitude of visual clutter would be above the threshold of reasonable good, i.e. I don't think it would be the sort of thing someone would say if I found someone on the street to show them this ad, and asked them if they thought it was reasonably good and think it's about as good as other ads.

For 2, for i) looking up Addidas ad on YouTube at least the first two results don't seem to have a single shot longer than 3 seconds before a cut:
https://www.youtube.com/watch?v=P1zOY2wjOQQ
https://www.youtube.com/watch?v=_R9uWNf7j6c
for ii) Certainly having high-level editing ability/judgment is a challenge, however, I for one find what was in that video compelling, and other than a few of the jump cuts having some broad consistency (e.g. in color) the overall level of coordination of the cuts and this does not really get in the way of it making for a compelling ad, so I don't think as a practical matter that presents much of a barrier.

Overall it seems like your reading of things requires a lot more autonomy of the system in producing the ad, which seems like a higher bar than would e.g. be applied to a human if I made an ad and used an existing song as the soundtrack I don't think anyone would object I did not make the ad.

I also read everything after "If an AI is specifically trained to not generate/be bad at ads but" as an alternative resolution criterion under that condition rather than details that inform the standard resolution criteria.

And have a nice day at work :)

I think the description of this is too vague. Please provide a prompt and criteria for evaluating the output so we can put this market to bed.

@quantizor Of course I won't provide a single prompt? That is not what the question is about, I am not going to modify the question a year and several hundred traders after opening it.
Some evaluation criteria:
- At least one major corporation is actually using the AI for their ads

-Several smaller corporations are using it
-No one is using it but I watch a bunch of ads produced by the AI and decide they're good enough.
As of now no model that I know of comes close to resolving this positively.

@vluzko Selling my stake then, good luck

@VincentLuczkow The criteria says "good at generating complex video with sound from relatively simple prompts" and "specifically prompting it to generate ads," and you said below, "it remains to be seen how cherry picked those are and whether they can do more than 3 seconds." Can I just confirm that this means the ad needs to be produced in a single prompt of the model? E.g., it cannot be many reasonably good 3-second video clips put together by a human curator with a soundtrack.

@Jacy Single prompt

Here's an actual ad for loo paper

https://www.instagram.com/reel/C0HzQQvrqot/?igshid=MTc4MmM1YmI2Ng==

@Dave_9000ish That's so creepy

@Dave_9000ish if psychos are the target demographic this resolves YES right /s

@TheBayesian wipe your mind, wipe your bum