I like to make long-term bets about future AI milestones. I've made 5 bets since 2013, and I've won 4 of them so far (the 5th one comes to term in 2026). Last month I won a 4-year bet I made with @alexandr_wang (CEO of Scale) about the deployment pace of self-driving taxis.
In Sept 2018, I bet that 4 years in the future, there'd be no publicly available fully-driverless taxi service in Silicon Valley. My thesis back then was that deployment in SV would start in 2023. I think Covid added ~6mo to that timeline -- today I'd see it in late 2023 or 2024.
Such a service already exists in in East Valley in Phoenix (Waymo One). Then there are two private beta ones where members of the public can apply: Cruise in SF and Waymo in downtown Phoenix and SF. I'm not aware of anything else.
At the same Sept 2018 party, I tried to hedge my first bet by finding someone to bet *against* driverless taxis being GA in SV in 2024. But I couldn't find anyone to take that bet. It seemed obvious to all AI folks back then that driverless taxis were coming within 2-3 years max.
Right now I only have one ongoing long-term bet, with @ChrSzegedy. If you're a mutual and you feel strongly about a future prediction that you know I disagree with, hit me up, I'm taking new bets. Timeline: late 2020s. 2030 max.
@lacker @fchollet @ChrSzegedy By implication (check else-thread), that his ARC challenge will fall before 2025 -- in about 2 years max.
@sir_deenicus @lacker @fchollet I wouldn't claim that
@ChrSzegedy @lacker @fchollet Hmm interesting. You think it'd be possible to have a system able to do non-trivial/interesting human scale maths and not be able to solve the puzzles?
@sir_deenicus @lacker @fchollet Possible. Math can be bootstrapped, IMO. For unspecified analogies it might be harder to design a curriculum to learn. Should check out ARC-2 in detail to make a more educated guess.