2 ms·

For the specific form of hallucination, which is called grounded factuality, we have trained a pretty good model that can detect if a claim is supported by a co

For the specific form of hallucination, which is called grounded factuality, we have trained a pretty good model that can detect if a claim is supported by a context. This is super useful for RAG. More info at https://bespokelabs.ai/bespoke-minicheck https://bespokelabs.ai/bespoke-minicheck.

mdaniel 2y ago
Your playground pre-populated example isn't doing you any favors, and the "examples" folder linked to on curator's GitHub would be better served by showing areas where your model shines, not "generate a poem" which hardly has any factuality to it. I don't have any earthly idea what camel.py is trying to showcase with respect to your model's capabilities I am open to the fact that maybe the value your service provides is in spitting out a percentage, even if it is - itself - hallucinated. But, hey, it's a metric that can be monitored