Another more practical issue with using LLMs for Zig is that it’s a quickly changing language, meaning LLMs may generate code for an older version of the language.
Yeah the author really should’ve taken some responsibility here. It’s true that the services they used have issues, but there’s plenty of blame to direct to themself
What % of users actually care that much about local LLMs? It appears to still be an inferior (though maybe decent) service compared to ChatGPT etc., and requires very top-end hardware. Is privacy _that_ important to people when their Google search history has been a gateway to the soul for years? I wonder if these machines would cost significantly less (or put the cost to other things, e.g. more CPU cores) without this emphasis on LLMs.
Why did Ladybird even attempt this with Swift, but (I presume) not with Rust? If they're going to go to the trouble of adding another language, does Rust not have a better history of C++ interop? Not to mention, Swift's GC doesn't seem great for the browser's performance.
I wonder what the internal conversations are like around memory safety at Apple right now. Do people feel comfortable enough with Swift's performance to replace key things like dyld and the OS? Are there specific asks in place for that to happen? Is Rust on the table? Or does C and C++ continue to dominate in these spaces?
If you're doing enough divisions with the same divisor, it'd be faster to do what compilers do for division by a known constant, where they multiply by an integer reciprocal and shift
I wonder if their "5.3" was continuously being updated, with regenerated benchmarks with each improvement, and they just stayed ready to release it when claude released
My experience with AI with its predecessor, Xcode 26.2, was _really_ bad. One bug made it objectively unusable, and there were lots of fun issues/huge functionality gaps on top of that. Apple doesn't really seem to "get" agent-based coding, but I'm curious to see the results of other braver souls with 26.3.
> It’s kind of like enabling LTO (Link-Time Optimization) across the libc boundary, except it’s done properly in the frontend instead of too late, in the linker
Why is the linker too late? Is Zig able to do optimizations in the frontend that, e.g., a linker working with LLVM IR is not?
This all just sounds like problems we see when making new features, of any sort, for customers. A feature is never objectively done, there are many opinions on its goodness or badness, once it’s released its mistakes can last with it, etc.
If this is a wicked problem, then so is much of other real-world engineering.
To be clear, is AI actually at play here, aside from the fact that the repo is for Gemini? It just looks like two simple rules that interact poorly, that we could've seen in 2015.
> Responses to my publication submissions often claimed such problems did not exist
I see this often even in communities of software engineers, where people who are unaware of certain limitations at scale will announce that the research is unnecessary