If we don't build AGI, the .govs and corpos of the world will build it first, and then use it against us.

Just saying

My manager uses Claude chatbot for writing all our user stories while we're on the call with him. It spits out so much shit! He'll say, "let's make sure it hasn't hallucinated anything .. nope"

There's no way he can be reading all that. I'll call out stuff early on, and he'll say "that's wrong, it hallucinated," and fix stuff and remove half of it, but 45 minutes into the story session, I think everyone tunes out.

Implementing this is going to be fucking hell.

Made this little example to show the difference in codegen quality using Gc smart pointers versus reference counted Rc's can make

Actually pretty crazy:

https://godbolt.org/z/vsTeGcz8M

I have small dream. It's going with my dad to shenzhen. Especially with my dad because he's quite tech save despite his age. He once was a programmer and that property seemed never wasted. He still understands a lot, including how (different) training of AI work on concept level.

In the Netherlands we have the datacenter days. I want to attend but the date is not known yet and you have to submit your presense quite some months up front for security reasons. Want to attend there as well. For me, it's magic how they pop up so fast. Would love to learn more about it from a person directly.

Anyone ever been in Shenzhen or something related? What is something a nerd should see?

Just read somebody say "she climbs like a homesick angel" to a fighter jet flying straight up and I think that's one of the most beautiful ways to describe it

Minecraft will yet again be changing how the version numbering works.

As a kid that always waited for major versions of the game to release and as a kid that identifies with specific versions of the game (1.6, 1.8 my BABIES), my heart aches and my confidence that the game has been fully ruined is continuing to grow.

At least I can just ignore all that and keep playing, the core mechanics are largely unchanged (except combat...)

A Nature Methods paper found that even containerized ML code can produce different results on different hardware. Same code + same data + different GPU = different outcome.

They are even seeing this with containers. Because hardware, drivers, and libraries in actual machine differ.

See people arguing over determinism in code gen. I wonder how much hardware matters.