While large language models are popular, relying solely on them for behavioral intent prediction is often flawed. Effective modeling requires more than just next-token prediction to accurately interpret complex human actions.