[dabockster]

LLMs in general are going to make C/C++ development viable again for the masses. A big reason everything went to web frameworks was memory safety - even big frameworks like Electron and React Native are memory safe overall. Writing low level code by hand was a migraine experience, even for experienced devs. AI changes this relationship entirely, even if you just use it as a pair programmer. I've had even small 8b models correctly call out memory unsafe code and suggest fixes. Larger models are even better.

Local LLMs? A new renaissance. All that power without having to pinky swear with a cloud provider that they won't just take your generated code and use it for themselves.

Expect to see some awesome Windows and Mac apps being developed in the coming months and years. 100% on device, memory safe, and with a thin resource footprint. The 1990s/2000s are coming back.

[cdaringe]

I hope so too. I’ve been using claude in some medium sized c projects and tbh its still not excellent. YMMV. Bolting it onto existing projects has been difficult. It doesn’t do a great job of understanding the build graph, nor the dependency relationships between the host system’s libraries (even with coaching, particularly in mac/arm space). It does pretty good at crushing impl code, but kind of the infrastructure and setup pieces are still lagging . I have a counter example on a smaller project, for which I was programming an E ink display, and it was able to understand all of the dependencies quite well, but that’s mainly cause they were localized and not heavily reliant on any sort of system libraries, and the build system was much simpler

[dabockster]

The Chinese models like Deepseek R1 and the Qwen Coder series have been way better at coding than any of the western models. And they're open source, too.

32 billion sized R1 and Qwen models blow GPT, Gemini, and Claude out of the water for coding tasks usually. And they can be made to run on a standard gaming PC off the shelf from somewhere like Best Buy or Costco. They will run really really slowly, though, but they will run.

[briansun]

Thanks-this is genuinely encouraging; I'd assumed AI help was strongest on front-end work(web apps/SwiftUI), so this is my first concrete example of an LLM catching memory‑unsafe C/C++-could you share your toolchain (CLI/IDE integration) and model details (name/quant/runtime), and what "awesome" 100% on‑device Windows/Mac apps you most want to see?

[dabockster]

I've been using a lot of the Chinese open source models like R1 and the Qwen coder series. I've also been trying some community mods of them from HuggingFace.

I was using a combo of Ollama + Roo Code for the front/back end but Ollama is kind of dumb when it comes to memory overload protections. I've since switched to LM Studio and it has a very annoying hard timeout of 2-3 minutes on its API server. Local isn't perfect right now, but when it does run you can see the potential as clear as day.