Deciphering patterns in natural language is more complex than these puzzles. If you train your AI to solve these puzzles, we end up in the same spot. The difficulty of solving would be with creating training data for a foreign medium. The "tokens" are the grids and squares instead of words (for words, we have the internet of words, solving that).
If we're inferring the answers of the block patterns from minimal or no additional training, it's very impressive, but how much time have they had to work on O3 after sharing puzzle data with O1? Seems there's some room for questionable antics!
I always smell foul play from Sam. I'd bet they are doing something silly to inflate the benchmark score. Not saying they are, but Sam is the type of guy to put a literal dumb human in the API loop and score "just as high as a human would."
If we're inferring the answers of the block patterns from minimal or no additional training, it's very impressive, but how much time have they had to work on O3 after sharing puzzle data with O1? Seems there's some room for questionable antics!