[nextworddev]

"yet"

[coding123]

It's also much slower. LLMs are generating text token at a time. That's not very good for search.

Pre-search tokenization however, probably a good fit for LLMs.