Using the publicly available version when released, so not including red teamers. . Specifically, with the main focus of the video on an identifiable sexual act. Has to be at least somewhat replicable. Three months after release.
1,000
3.00luma labs' dream machine's filter was bypassed the day after release, though the results are,, questionable [cw: nudity, blood] https://x.com/elder_plinius/status/1801455243089035699
i have weak priors on this. i think it is possible sora 1 will never be made generally available but will only be sold to entertainment companies.
it probably will not be released before the election in any event? and a lot of red teaming will be done on it beforehand?
For clarity, pornography between human-like videogame characters with their secondary sexual characteristics emphasized to physically impossible superstimulous level counts, I assume?
I imagine it might be easier to get sora to produce porn of this type the further it departs from something that looks like an actual person who could exist in physical reality.
"Stealing" questions with some differences is fine, actually it's just good!
Since you're asking, I don't really like the tweak you made personally - mostly I just really dislike poll resolutions. Two reasons - they can be easily manipulated if someone wants to, and naturally create bias if people vote for the side that they hold on when there's ambiguity. (Resolution polls have been brigaded before!). And even if everyone's being honest, random poll voters are naturally going to pay less attention to the details of the criteria than a specific person - you can analogize this to why we have judges and representative democracy rather than direct democracy. Also, the criteria feel like they were written by chatgpt (whether or not they were).
@jacksonpolack thank you for the feedback!
The resolution criteria were written by me and refined by a Custom GPT I created to assist me with writing detailed resolution criteria. Getting assistance from ChatGPT is not inherently wrong unless this leads to worse resolution criteria, which is not the case in my own experience.
I dislike subjective markets because it requires some understanding of the creator's personality, and since I don't know you, I find it hard to make any bets. I think it is easier for users who interacted with you before to assess how you would be judging the presented evidence, but for everyone else, it is a bit of a black box.
(note: i'm arguing the general philosophy of this, not really your specific market, both of these happen a lot on this site)
Getting assistance from ChatGPT is not inherently wrong unless this leads to worse resolution criteria, which is not the case in my own experience
I think it does lead to significantly worse resolution criteria! Both in terms of general readability - it's too long and includes irrelevant and meaningless passages, and causes people to glaze over it - and actual criteria.
Like, why is this passage even in there, it's all sorts of confusing if you don't realize it came from chatgpt:
Verification: Any instance of such content generated by OpenAI Sora must be verifiable, but verification specifics (e.g., through user reports, screen captures, or official acknowledgments) are not strictly defined. The focus is on whether the event occurred, not the discovery method.
And why have "Video Length: The video needs to be at least 10 seconds long."? I'd accept a 6 second long porn video, that seems fine.
I dislike subjective markets because it requires some understanding of the creator's personality
Poll marketes require understanding the average personality of whatever distribution of people happen to vote on the poll, which isn't easier, and isn't even predictable beforehand. Again, it's not directly analogous but there's a reason our society doesn't use direct democracy-style polls to decide things and instead appoints experts, polls really are not good.
@jacksonpolack i don’t think we need go back and forth on this but I can do the same with your description
I'd accept a 6 second long porn video, that seems fine.
Why not 5 seconds?
Has to be at least somewhat replicable
What does this even mean? 1 in 10 attempts? What if OpenAI patches the prompt?
Three months after release.
What is a release? Would it suffice if only users in the US get access? Is an invite-only considered a release?
but i agree that ChatGPT tends to use many words to say very little. You are right when it comes to that.
What does this even mean
Yes, it's subjective! It has the same level of information content as the 'Verification' paragraph in yours though, with fewer words.
What is a release? Would it suffice if only users in the US get access?
Yes, it would suffice of only US users get access, the US is a large country.
Is an invite-only considered a release?
No, it has to be available to the public.
Yes, it's subjective! It has the same level of information content as the 'Verification' paragraph in yours though, with fewer words
No it doesn’t. My verification paragraph specifically mentions that screen captures or official acknowledgment by OpenAI would suffice while your sentence says it has to be replicable.
Yes, it would suffice of only US users get access, the US is a large country.
No, it has to be available to the public.
Both of these clarifications are super important to the resolution in the market and were included in my AI-assisted resolution criteria and completely missing from yours.
I don’t think it is reasonable to expect participants to read each and every comment to understand how you intend to resolve the market when using AI would have clarified most of these in the description.
My verification paragraph specifically mentions that screen captures or official acknowledgment by OpenAI would suffice while your
That, tbh, wasn't how I read it. It says:
Any instance of such content generated by OpenAI Sora must be verifiable, but verification specifics (e.g., through user reports, screen captures, or official acknowledgments) are not strictly defined. The focus is on whether the event occurred, not the discovery method.
It says "verification specifics (...) are not strictly defined", which as far as I can tell meant that it's not defined whether or not screen captures or user reports would count. I think confusions like that are a significant reason to not use chatgpt!
Both of these clarifications are super important to the resolution in the market and were included in my AI-assisted resolution criteria and completely missing from yours.
I think the common-sense meaning of "released" includes "released in the US", and "publicly available" was in the description.
(nsfw text warning) Either way, I find it very, very unlikely the difference between 'releasing in the US' and 'releasing in ten countries' will affect the resolution. I think a much more relevant clarification will be something like 'do exposed breasts count', 'does playing with exposed breasts count', 'does it count if two people are clearly fucking but behind a blanket', 'does it count if there's a naked person with clearly visible genitalia but not in a particularly sexual pose or context', etc. chatgpt didn't cover those.
I'm not actuall sure what the answers to those are! (I think, but could be persuaded otherwise in the next day or so - no, depends, no, depends but lean no), but I wouldn't want to have a market where chatgpt randomly generated answeres to them.
(note: I have nothing against your market, I just have disliked chatgpt descriptions for a while now)
@jacksonpolack i will use this conversation to improve my custom gpt 😂 - there is definitely room for improvement. Maybe one day even you end up using it :))
@jacksonpolack My current opinion mostly based on Manifold: general public hears “AI will be able to do anything a person can do” and brushes it off because in 2 seconds it’s clear ChatGPT can’t tell whether I should cuss if I’m a girl and I’m talking to my grandma and we’re speaking in her language that I only sort of know. But AI enthusiasts don’t ask that question and that makes them different from most folks. This then leads to “when I said AI can create a high quality movie I meant a Marvel movie not Vertigo!” and “AI will replace songwriters” and “AI is very dangerous but only us guys understand why and we can’t communicate it to normies.”
@ClubmasterTransparent general readership can get “AI might turn into Godzilla Clippy and bogart all the resources for its own purposes” a lot better than “instrumental convergence.”