undefined | Better HN

0 pointszackangelo1y ago0 comments

Would you care to share your prompts?

They posted a haystack benchmark in the blog post that seems too good to be true.

0 comments

2 comments · 2 top-level

I wasn't scientific about it, unfortunately. My searches were natural language, not token-based, though.

yeah, when i saw that they have 100% coverage with 1M token, i thought this must be a placeholder image, for when the actual results come in.

Because there is no variation, nothing.

j / k navigate · click thread line to collapse