House of AI Cards
Tech CEOs promise superintelligent AI by the 2030s, but recent tests reported by New Scientist suggest caution. OpenAI’s GPT-4.5 showed limited gains despite massive investment. Researchers at Apple found AI reasoning models fail logic puzzles like Tower of Hanoi (see below), performing worse as complexity increases. University of Maryland studies revealed longer "chain-of-thought" processes lowered mathematical accuracy. “Fundamentally, there is a mismatch between what these models are trained to do, which is next-word prediction, as opposed to what we are trying to get them to do, which is to produce reasoning,” says Univ. of Cambridge researcher Andreas Vlachos. OpenAI disagrees, claiming new architectures like o1 will keep scaling progress alive. As noted in Is It Thinking?, reasoning remains AI’s blind spot. Read more in New Scientist (paywall)
