1 paper · 1 filter
Francois Crespin, Fabian M. Suchanek, Nils Holzenberger
We propose a new method that allows an LLM to automatically pull in factual knowledge from a knowledge base during token generation. This means that (1)~factual knowledge in the LL…