(ashishb.net)

33 points ashishb | 2 comments | 08 Mar 25 05:14 UTC | HN request time: 0.421s | source

Show context

darthrupert ◴[08 Mar 25 07:04 UTC] No.43298148[source]▶

Now that Rust exists and llm can easily lift a more junior person over the initially difficult parts, there are not many things you should use python for.

replies(1): >>43298193 #

randomNumber7 ◴[08 Mar 25 07:10 UTC] No.43298193[source]▶

>>43298148 #

Now that AI based verification exists, I can just write C again and call it a day.

replies(2): >>43298504 #>>43298947 #

1. timeon ◴[08 Mar 25 10:06 UTC] No.43298947[source]▶

>>43298193 #

I wonder if Rust or C would decrease need for power use while training those models.

replies(1): >>43329719 #

2. ashishb ◴[11 Mar 25 06:27 UTC] No.43329719[source]▶

>>43298947 (TP) #

It definitely will. Whether it will make a noticable difference depends on how much non-cuda work happens in model training.

↑

It is hard to recommend Python in production