The series of puzzles, which are relatively easy for humans but will challenge today’s most sophisticated artificial intelligence models, is intended to inspire AI developers to develop new techniques, and any AI that solves the puzzles will earn its creator a share of a $1 million prize.
Companies like OpenAI are already claiming that AI models like GPT-4 have demonstrated “human-level performance” on real-world tests like college entrance exams and the bar exam, but that this isn’t because the models reason intelligently like humans do. Mike Knoop in…
Source: www.newscientist.com