Proboscis_Monkey's blog

By Proboscis_Monkey, history, 3 months ago, In English

My question is, how is that revolution possible? We are familiar with chat GPT or Bard AI, which are hyped. However, they are unable to solve even Codeforces A (of div2), which is considered normal. But how will the Alpha code be able to solve problems like an expert or a CM coder? Will it be able to surpass the skills of other lower-rated coders in real-time contests?

  • Vote: I like it
  • -17
  • Vote: I do not like it

»
3 months ago, # |
Rev. 3   Vote: I like it +5 Vote: I do not like it

It's not solving them like an expert or cm. It uses a big database with 30million solutions and tries to find which solution fits the given problem.

I don't know how much logic it is doing, but it seems not much. The second highest rated problem it solved is 2100. And you have to be aware that this was not done under contest rules:

  • the AI has virtually infinite time (AIs are easily faster than humans)
  • the AI is cheating by hard coding test cases to see more example input/output.
  • the Google paper is not convincing in that the AI is not cheating by looking up other people's solutions.