# @voooooogel — 2023-11-13

♥2 ↻0 · https://x.com/voooooogel/status/1723945803792666894

then during inference, take the question, have the model hallucinate a chunk based on it, then retrieve the real chunk closest to the hallucinated chunk, sub that in for the hallucinated one, and finally generate the (now grounded) answer https://t.co/aoiIkY9664

tags: author:voooooogel, kind:tweet, thread-context, year:2023