Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

But just 2 months ago there was an article on HN that stated:

The database is a key value store prepopulated with the english names of countries and their capital cities from the continent of Europe. Selecting the country will perform a search of the matching capital. On a 2019 Macbook Pro laptop, the example searches take under 80 seconds.

https://news.ycombinator.com/item?id=23435305

And today this article claims only a 40x compute cost for "machine learning"?

What is the cause of the disparity?



I think the 40x overhead is a case of comparing throughput overhead (from what I know, FHE based secure inference protocols have poor latency, but can process many predictions in parallel, improving throughput)


My layman guess is that the FME penalty goes up exponentially to the complexity of an operation.


Doing a exact string match on 200-ish rows in 80 seconds on a modern computer is so inefficient that I have a hard time seeing any less complex but useful operations whatsoever. Perhaps I'm just not clever enough, but for now homomorphic encryption seems like it isn't useful for common, real world usecases to me.


This isn’t true, it’s just that different kinds of operations are more or less efficient in FHE




Consider applying for YC's Winter 2026 batch! Applications are open till Nov 10

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: