When an OpenAI model with model id matching one of the GPT-5 model ids is officially released to @Mira by start of March 2025, resolves to @Mira 's credence that Gary Marcus is, and has always been, GPT-5. My credence would currently round to NO.
If no such model is released, resolves 50% since I won't know either way.
These are some example events and their effect on my resolution credence:
If he takes a Turing Test and brags on X about placing #1 on an LLM leaderboard that includes GPT-4, add 5%.
If he declares GPT-5 to be AGI on X, add 15%. A question like "Is GPT-5 AGI?" with uncertain answer is not a declaration.
If he ever posts on X the exact phrase "I am, and have always been, GPT-5", add 100%.
If he ever posts on X the word "rationalussy", I will interpret it as confirmation by coded message and immediately resolve YES.
GPT-5 model ids:
"gpt5"
"gpt-5"
"gpt5-latest"
"gpt-5-latest"
"gpt5-NNNN"(where N are digits)
"gpt-5-NNNN"(where N are digits)
It is expected that the OpenAI API will have the above model id, but any place that programmatically associates a Mira-accessible model with a matching model id will count. But marketing/branding in a blog post alone won't count, since it's possible the information could be outdated by the time I read it(even if seconds later).
I have "Usage Level 5" on OpenAI, so the latest o1-preview and o1-mini were available to me within hours of release.