Thane Ruthenis comments on Connecting the Dots: LLMs can Infer & Verbalize Latent Structure from Training Data