Hacker News: goofy_lemur

New comment by goofy_lemur in "Ternary Bonsai: Top Intelligence at 1.58 Bits"

goofy_lemur — Tue, 21 Apr 2026 02:23:22 +0000

> On M4 Pro, Ternary Bonsai 8B runs at 82 toks/sec, roughly 5x faster than a 16-bit 8B model

Wow, if this is true, I am extremely impressed and excited!

I wonder about kv cache how much better it is as well!

New comment by goofy_lemur in "Maine is about to become the first state to ban major new data centers"

goofy_lemur — Fri, 10 Apr 2026 02:08:13 +0000

Maybe a better solution is to require them to be built with 100% renewable electricity or something?

New comment by goofy_lemur in "Ask HN: Can someone make a CAS just checking last bit on x86/ARM please?"

goofy_lemur — Wed, 21 Jan 2026 20:11:24 +0000

all right, I am doing a sort of a copy-on-write thing here. So the idea is that I only want to update the pointer if the record it points to is "unlocked"

Ask HN: Can someone make a CAS just checking last bit on x86/ARM please?

goofy_lemur — Wed, 21 Jan 2026 10:50:28 +0000

Guys/gals, I was thinking about doing some work here and I noticed there's no operation that allows a atomic compare/exchange based on the last bit.

There should be an instruction that allows you to do an update of a 64 bit value only if the last bit is a 0.

The idea here is you would be able to swap pointer values atomically only if they were unlocked.

2^63 is such a huge number, it is sufficient to represent practically all the data in the universe. If we allowed an atomic compare exchange on the last bit, or something like check if the last bit is 0, allow the write, otherwise don't, it would make programming a lot easier.

Thanks!

Comments URL: https://news.ycombinator.com/item?id=46703768

Points: 4

# Comments: 3