Did you read the study? It's hilarious. They're using LLMs to "grade" the number of observed "skills" based on the output of LLMs. They're using a stochastic parrot to evaluate another stochastic parrot, and concluding that there is some kind of emergent "skill" going on. Sheeeesh. It'd assume the authors of the paper are just having a laugh. But, one thing is for sure, the AI stupidity train keeps chugging along.