3 ms·
That's not entirely accurate. LLMs encode some level of understanding of their training set. Whether that's sufficient for a specific purpose, or sufficiently
by ethbr1 3y ago
That's not entirely accurate.
LLMs encode some level of understanding of their training set.
Whether that's sufficient for a specific purpose, or sufficiently comprehensive to generate side effects, is an open question.
* Caveat: with regards to introspection, this also assumes it's not specifically guarded against and opaquely lying.