Resolution criteria: when an AI model does contests on codeforces and gets a rating higher than my all time high, and it stays that way for more than a month this market resolves 100% to the quarter it got that rank. (so, if a model comes out 20 december, and does a lot of contests 23, 24, 25 and 26 december 2023, and by dec 27 has higher rating than my ATH, and then its rating stays above mine through jan 27 2024, the market resolves Q4 2023)
Context:
Alpha code 2 was released
https://news.ycombinator.com/item?id=38544935
https://codeforces.com/profile/AdamantChicken2
https://codeforces.com/blog/entry/99566
I'm currently ~1700 rated on codeforces. Which should be slightly better than what AlphaCode 2 does. According to the report it is better than 85% of participants And from what I gather 1700 is better than 89.7% of participants. The resolution criteria means I can panic and do a lot of contests if results come out that imply resolution, and then maybe do better, to push out the date at which the market resolves. But obviously, if a model comes out that is grandmaster level, that would be nigh impossible to do.
1,000
3.00Just to be clear. If anyone is wondering
1) GPT o1 (and the fine tuned version on code) would NOT cause yes resolution, because too low rating.
2) If it DID have high enough rating, it'd cause the market to resolve Q3 2024
> This model far exceeded both GPT-4o and o1—it achieved an Elo rating of 1807, performing better than 93% of competitors.
I'm currently 1900, so I am quite confident. alphacode2 couldn't cause yes resolution.
Edit: Also updated what I realized might've been an ambiguity in the resolution criteria. Resolution is when models are released / published. Like if someone got access to alphacode2 now, and did tests satisfying the criteria in the description, it'd be Q3 2023 resolution still, even though we're in Q2 2024.