> that language models have no problem reliably inferring

... the article did give me a different impression.