In this #Bluesky thread, C.T. Bergstrom notes that some scientists expect #LLMs to become highly effective individualized tutors. However, in a simple example he demonstrates that #ChatGPT-5 is far from achieving that. It quickly becomes confused by the evolving context of an ongoing chat and fails to answer questions it could solve at the beginning of a new session.
Raises an important point about current #LLM: context awareness alone does not equal understanding.
馃實 https://bsky.app/profile/carlbergstrom.com/post/3m4frfj4fok2e
