2,258 karma · joined November 6, 2021
Whenever I try to do something with ChatGPT or Gemini, it just refuses to do things citing limitations in its web browsing capabilities. An everyday LLM that can actually fill forms and do stuff for you on the Internet without having to run on your computer, write programs or interact through special harnesses is what normal people actually need.
It's a different need from the agentic coding type of LLM, where you provide the environment where it can write and execute arbitrary code.
Unfortunately as the crow flies that seems to be closer to 800 miles than 125.
I'm in London and the airport itself is already 60 miles away.
With macOS I have to reverse engineer their bootloader with radare2 and remove a stupid safety check just to be able to load a driver.
And even if everyone is using the same LLM model, quality varies on lot on who's driving it.
Windows emulation did see some improvements with Valve's sponsorship, but I wouldn't really consider that significant, and it's not like mac has any comparable technology anyway.
In general, the model on Mac is that you need to run latest. It's not like you're gonna use it for a production server, they're meant to be personal devices, so upgrading them is not a major issue.
I reported it about ten times, Apple never replied or addressed it. After a while I decided to take things into my own hands, and to write my own driver.
Just installing a VM you can load a driver in for development and testing was a two-week effort. Everything locked down. Their VM infrastructure is so bad you can't even properly emulate or route the device, so in the end the only way to test it properly was to install the driver in the real thing, which is extremely disturbing to productivity, as the device you're working on constantly crashes.
The whole platform is developer-hostile, which is essentially the same thing as user-hostile as far as I'm concerned.
Meanwhile, on Linux, zero issues, and writing or testing a driver is child's play. It's literally built for tinkerers.
I think what changed is that you got better at Linux. Leveled-up, so to speak.
No surprise here.
the tell-tale mark of AI code is highly over-engineered local solutions to trivial problems that don't matter, or that were already solved better elsewhere and that no sane human would ever duplicate.
It generally performs better if the tasks are broken done into small manageable pieces, and the person is actually reviewing and calling out problems, which usually requires the person to be a competent engineer in the problem domain to begin with.
But yes, I have personally used it to build what the OP calls a JIT. I would usually write that by hand and it would take me one week. The AI does it in an hour.
By not using LLVM, you're missing all the optimizations it does.
Why would anyone expect big broad changes? And historically, whenever they did happen anyway through forced compromise, they resulted in failure because they were not consistently implemented.
The only way that works is small compatible and iterative changes.
Old Street, walk-in only.
It didn't work in practice, because people want to structure their document based on how they expect it to be presented.
My understanding is thay HTML5 was built on the understanding that language purity just went in the way of productive compromise.
It remains quite different from a genuine ramen shop in Japan.
Your code shouldn't need to make assumption about whether exception are being thrown or not.
Your mistake is thinking that a function potentially throwing means you need to catch it.