Summary
This market will track the movement of Manifold's most popular near-term AGI market after GPT-5 is formally released. This market will compare the odds of AI beating the Longbets version of the Turing test by the end of 2029 before GPT-5 is released with the odds a month afterwards. If the odds have risen, this market resolves to Yes. Otherwise, this market resolves to No.
Fine print:
AGI: Artificial general intelligence lacks a shared definition. I am using the Longbets Turing test because it essentially measures whether AI is as "smart" as humans in nearly every domain (which is a decent proxy for most definitions of AGI), and it is Manifold's most popular market measuring short-term timelines.
Resolution criteria: Any amount higher (including fractions of a percent) will be admitted as proof of the market rising.
Resolution timing: To avoid confusion about when in the day GPT-5 should be considered to be released, I will use the status of the market at midnight before it is unveiled. For example, if Sam Altman gets on stage to show off GPT-5 at 13:00 on March 5, I will use the price of the x-risk market from midnight 13 hours earlier as a baseline.
To avoid last-minute rigging of the outcome, I will resolve this market after around a month at an undisclosed time. As usual, I will not bet in here to avoid conflict of interest. With that said, I can't guarantee somebody won't dump 1000 mana into a Yes position for the X-risk market with the explicit purpose of manipulating this market. Manifold is the Wild West of prediction markets, and I'd recommend betting accordingly.
Gpt-5: There is a chance that OpenAI's next flagship model will have a different name. If there are reasonable questions over whether a release is "GPT-5 with a different name" I will use a manifold poll to resolve this.
Launch: I will consider a launch to be a formal unveiling of capabilities. Rumors/leaks don't count. It will need to meet the following conditions
I'll expect a presentation with Sam and/or Satya, or at the very least a paper talking about all the benchmarks it crushes.
Access given to the public, or at least journalists/privileged outsiders. We need somebody who doesn't work at OpenAI or Microsoft able to use the model and give their honest opinion to the public.