AGI Can Be Safe
Data Skeptic
00:00
The Relationship Between LLMs and the MDP Paradigm
large language models have been trained on a huge corpus of the internet which contains facts and fiction. People are still trying to show hold them a bit into the MDP paradigm by saying okay um a last language model you could say if you're sure if you put it in the MDP principally the last language model is a predictor of future sentences, says Kraxti. The dream of AR researchers many ML researchers is that they can somehow tweak these large language models so that what comes out will always be true but nowhere near enough where these things are actually useful yet.
Transcript
Play full episode
