Chad Smith

SWE at meta, love creating things and solving problems. Previously worked in aerospace on safety critical software/GNC.

[Chad Smith @cs01_software]

@alpeh_v

[Chad Smith @cs01_software]

@schteppe https://t.co/B9EW5hL39i

[Chad Smith @cs01_software]

@damian_b I agree that safety and shifting errors left is very important for AI. I think we are nearly able to design languages on the fly. A language needs a compiler. And these days a compiler is basically a tool that generates LLVM IR. AI knows how to write LLVM IR very well.

[Chad Smith @cs01_software]

@haider1 That's ironic because c++ has been bloated and unsafe long before AI came around.

[Chad Smith @cs01_software]

@nearX86 @ThePrimeagen https://t.co/DLni1vObzY

[Chad Smith @cs01_software]

@nearX86 @ThePrimeagen Maybe not the best term I'll rename. It measures Halstead Volume (total information content: operators × operands), not conceptual difficulty. Zig scores high because it's verbose, not because it's harder to reason about. It's actually highest of all! https://t.co/g2hWJhYMAR

[Chad Smith @cs01_software]

@ThePrimeagen There is little objectivity around programming language debates, and AI will bring an explosion of new ones (I already made two). I took a shot at a more formal way to compare them with Language Explorer.

https://t.co/USmtdx5mya

[Chad Smith @cs01_software]

@ThePrimeagen magpie seems to be much more ai-first, thought I couldn't get it to work when I tried it

https://t.co/wimpdUbMq4

[Chad Smith @cs01_software]

@jsep_do @ctatedev This one seemed much more agent-first https://t.co/9F3I3lLcMq

[Daniel Hong 🪄 @unifiedh]

I built the world's first programming language designed for LLMs and AI Agents.

Introducing Magpie.

Check it out yourself at: https://t.co/4bVeYmLw0I

Details below.

[Chad Smith @cs01_software]

@trq212 You can just imagine the language you want and will it into a compiler

C++ null dereference warnings in clang? Ok. https://t.co/9MRZ0BUSln

Typescript frontend with LLVM backend? Ok. https://t.co/EuSzrmYQXq

[Chad Smith @cs01_software]

@RhysSullivan @jarredsumner It has been done already https://t.co/5F4ihjJ16T @realamlug

[Chad Smith @cs01_software]

@aksiksi @glcst I know people who actively work on extremely complex c++ codebases, that have a large number of runtime crashes, and still prefer it over rust. I really don't get that mindset.

[Chad Smith @cs01_software]

@jarredsumner I had Claude segfault before. I assumed it was something I did wrong for some reason but I guess it was legitimate.

[Chad Smith @cs01_software]

@jarredsumner @mitsuhiko Cut out the middleman and write a TS/bun to llvm compiler. Who needs zig anymore?

[Chad Smith @cs01_software]

@headinthebox Opus 4.7 is particularly bad at it (along with everything else). I tell it to estimate in terms of LOC instead of time and it helps, though it still sometimes does timeframes.

[Chad Smith @cs01_software]

@bcherny Opus 4.7 is really bad, even after the post-mortem. I am going back to 4.6. Please do not remove 4.6 as an option, I will switch to Codex immediately.

[Chad Smith @cs01_software]

@xwanyex Again, never could have done this so quickly and so thoroughly without LLMs.

[Chad Smith @cs01_software]

@xwanyex Try it in the online playground to compare against stock clang, and clang analyzer. https://t.co/eHwcv06jB3