@ShiboChenTech And there is absolutely contribution from the developer community. They optimize, quantize, and fix bugs in the models. Hope open weight models win out in the end.
@ShiboChenTech Haha. Yeah I am excited for those as well. Just wish I could get Linux with a MacBook hardware. Linux laptops always seems to be the worst of all worlds (bad firmware with mediocre hardware)
@ShiboChenTech If anything I think I think the Mac is the best part of the Apple Ecosystem because there is nothing even comparable on the hardware side. It is at least Unix, even if it doesn’t give the full control of Linux.
@code_report@mattgodbolt It is interesting to see languages like zig where compiles times are a first class citizen and languages like C++ and Rust where it is an afterthought. Zig will change or not implement features if they have too much compile time impact.
@lemire It would be interesting to see if there were productively numbers form when the internet was invented. I feel like these new technologies change things very little, then all at once.
@laurencetratt I had an LLM tell me about this just the other day. I was trying to benchmark the overhead of running a command and it suggested this one because it does nothing. Worked great.
@asmeurer@Steve_Yegge Or similarly, they will often write large complex regex when asked to parse something. They work, but are incomprehensible. You need to tell it explicitly how to tokenize or parse.
@ChadNauseam If you look at the fastest typists in the world, they mostly use QWERTY. If it mattered, you would see a better performance from alternative layouts.
@intellectronica@mitsuhiko I don’t see a reason not to use the loss leader while it is cheaper. You can always switch to a different model or provider later.