closes
Dec 31, 2023Will AI outcompete best humans in competitive programming before the end of 2023?
5%
chance
DeepMind has recently published a pre-print stating that their AlphaCode AI is as good as a median human competitor in competitive programming. See https://deepmind.com/blog/article/Competitive-programming-with-AlphaCode . Will DeepMind, or anyone else provide evidence in 2023 they can beat the best human competitors?
#AI #DeepMind
Extraordinarily unlikely before 2025.
Any legitimate problem takes perfect understanding of the prompt and then limits test cases (e.g. 60s delay between submissions that come in under 3 minutes)
2037: probably
0
You can only really do this by constraining the problem space to something that your 'AI' is good at. AlphaCode reads like that. Vastly overspecced, vastly abstract and ridiculously verbose. It's not a real world thing.
0
The median competitive programmer is a low bar, contests typically have a large number of contestants who do not know what they are doing.
I'm confident we'll see improvement here over the next year, from both tech improvements and OpenAI finding clever ways to augment their network with brute-force, but the long tail of competitive programming problems requires more consistent reading skills than GPT-3 has demonstrated so far. To get to near-perfect consistency on medium-difficulty problems (which is what top human programmers achieve) in two years would imply very short AGI timelines - so not impossible, but 26% seems overpriced.
0
Play-money trading
Mana (M$) is the play-money used by our platform to keep track of your trades. It's completely free for you and your friends to get started!
