“Amjad Masad points out that in a recent test, three out of four major LLMs failed a simple character-counting question, and the one that succeeded required 15 seconds of processing time, illustrating their current limitations.”