Words can have multiple meanings. Feet can be the things on the end of legs, or a unit of measurement.
LLMs process data in a more intelligent manner than previous systems. The solution, whatever it is, is presented in a thoughtful manner to the system operator. "Thinking" seems like a pretty convenient term for the process. It doesn't have to imply sentience or sapience.
Its the same with "Learning" or "Training" these models aren't actually learning, and they certainly aren't being trained. But these words are convenient shorthand that conveys a similar process.
The battle for language prescriptivists isn't just to ban a certain use of a word, its to find a convenient alternative. I dont see any here.