4 ms·
Currently, LLMs generally don't give an indication of confidence level when generating output. But could they? Is it technically plausible for a LLM to "know"
by ipsento606 2y ago
Currently, LLMs generally don't give an indication of confidence level when generating output.
But could they? Is it technically plausible for a LLM to "know" when the extrapolations made are more vs. less tenuous?
- segmondy 2y agoSolved via RAG, in the future it might be possible to encode it in the weights and via architecture design.
- flohofwoe 2y agoSources would be much more important IMHO. Give me a list of source links which the output is a remix of. Not possible? Well, then back to the drawing board and figure out a solution you AI experts!