"First AGI" refers to the initial publicly acknowledged AI that surpasses the Longbets Turing test and becomes accessible to everyone (in at least 3 countries), prompting the resolution of the "AGI When?" market of the Manifold AI countdown (if not yet resolved).
Resolution Criteria:
Multiple responses can be resolved as YES.
Additional responses can be submitted.
If the first AI that meets these criteria is classified as ASI rather than just AGI, it will still qualify under this market.
An AI that has passed the Longbets Turing test but is not publicly accessible will not be considered the "first." It only counts as the "first" when it becomes accessible to everyone. If another AGI is later announced, which also meets these criteria but is inferior, it will still be considered the "first" if it becomes accessible first.
The term "first AGI/ASI" is used in a strict sense. Subsequent iterations or finetuned versions of the same model or architecture post-initial public release do not qualify. Only the initial public and accessible version that passes the Longbets Turing test and is officially announced counts.
All responses pertain to this initial iteration, even if superior models are released later. Accessibility includes average individuals being able to use the AI, whether for free or through paid means like an API.
Waitlists for access count as "already accessible by everyone" as long as they do not prioritize individuals based on something other than join order.
Once all specified conditions are met, I will begin resolving the eligible responses. Some responses might require direct access to the AGI and could take longer to resolve.
If multiple accessible AGI versions are released simultaneously, the one to first publish its Longbets Turing test results will be the benchmark for resolving the market. If all versions publish their results concurrently and meet the criteria, each applicable response will be resolved in favor of a YES (e.g., "Has less than 1 trillion parameters" and "Has more than 1 trillion parameters" and there are two models, one with less and another with more, both resolve to YES).
If the "AGI When?" market is unresolved at the time of the AGI release but resolves later, it will count.
If for any reason Manifold deletes, cancels, or closes the "AGI When?" market before an AGI beats the Longbets Turing test, then the requirement of that market resolution for this to resolve won't be necessary.
I will answer doubts about responses added by me. Doubts about responses added by others must be answered by them.
Responses that are ambiguous or can't be immediately resolved will remain open until there is sufficient clarity. In the absence of a clear resolution for a response, I will resolve to my best criteria, asking for feedback in comments or making a poll about it.
Responses deemed invalid or redundant will be resolved as NO.
Before this accessible first AGI, I won't resolve any response (with the exception of invalid or redundant ones), even if the resolution is clear.
After 22 March 2050, if no AGI meets these criteria, all responses will be resolved as N/A by a moderator (with the exception of the response "This market will resolve N/A" that will be resolved as YES in that case or as NO in any other case).
Additional context for resolution criteria of your added response can be provided in comments (tagging the response), which I will consider during resolution. This should not be used to unfairly alter the resolution criteria or mislead traders.
Repeated or similar themes in responses are permissible if they clearly differ in their criteria. However, they should not be abused to crowd the market with nearly identical questions.
Responses must be made so that they cannot be resolved before the first AGI. So "Released before 2025" is not valid as it could be resolved as NO after 2025 ends before the first AGI, but "Started training in 2025" is valid.
Responses must be about this first AGI (development, characteristics, etc.) or consequences of it (mass unemployment, protests, etc.).
Anything that implies a time interval after this first AGI (like "80% of global population loses its job"), unless otherwise stated, will assume a 1-year wait to resolve.
If an important clause of resolution criteria fits inside the response text (without losing details), I encourage you to write it there instead of in comments. This is very useful to make a response with a very different criteria from the default I use.
OP Trading: Even if some responses are objective, others may not be, and since I would have the last word in ambiguous responses, I won't bet on this market.