Gamedev, but that was long time ago. Nowadays, I'm more curious about how to solve problems.

It was a great book. Especially later on and the parts about how we built modern economics. A lot of the details were brutal but really eye-opening.

Great. Just another Verschwörungstheoretiker. L

What Weinstein refers to as "Lagrangian" is incomplete and depending on the undefined SHIAB operator. In other words - useless. youtu.be/jz7Trp5rTOY?...

She's a clickbaiter. All she cares about are YT money, let's not kid ourselves.

I only learned about this just now. Respect to you Sean for doing this.

It all started with this recent "private posts leak" in #pixelfed as described here: fokus.cool/2025/03/25/p...

Private posts meant only to a Mastodon account's followers were visible in Pixelfed. This is bad.

Looks majestic! I hope it all goes well. 🤞

My current concerns are price and time. Can I vibe a project and have it cost-effective in terms of end-price ($ for credits) and invest less dev time. I completely discard #vibecode for huge codebases with frequent features and lots of collaborators. I don’t see this happening unless modularized.

I don't think 1-5 t/sec is an option. Your point is good though, bigger models are in fact quite capable. Like I said my point is to find out if there could be a (compromise) solution for on premises scenarios. Thanks.

True. I'm thinking an on premises product, so I can't expect top hdw from customers, and so I'm trying to think of min. required hdw for inference i.e. using fine-tuned models trained on data that covers my niche use case. However, even with smaller params, inference can be quite resource hungry.