Anthropic's Claude 3.7 Sonnet is the new king 
of code generation (but only with help), and DeepSeek R1 disappoints https://symflower.com/en/company/blog/2025/dev-quality-eval-v1.0-anthropic-s-claude-3.7-sonnet-is-the-king-with-help-and-deepseek-r1-disappoints/
#Java #CodeGen #genai #llm
Somebody reviewed Deepcoder 14B vs Qwen2.5-coder-32B. Qwen2.5-coder 32B seems to be still ahead, over 6 months after release. I have high expectations for Qwen3-coder, we will see.
https://www.reddit.com/r/LocalLLaMA/comments/1jwhp26/deepcoder_14b_vs_qwen25_coder_32b_vs_qwq_32b/
Switched UI to #Python #Textual and reached #LLM #codegen limits. Did some fixes to layout by hand to make it work on 80x24 and left notes for achieving more advanced UI features possibly later https://codeberg.org/jasalt/espeak-variator
Comparing Use of AI vs Learning to Code Has Nothing to Do With Knowledge https://christiantietze.de/posts/2025/02/using-ai-vs-learning-to-code-is-diminishing-knowledge-by-degrees/
In its extreme form, LLM usage has nothing to do with gaining knowledge at all.
So speed-vs-knowledge comparisons are actually wrong.
Ah, the fun of doing spec-driven development with infinite interns:
My LLM codegen workflow atm https://harper.blog/2025/02/16/my-llm-codegen-workflow-atm/
#software #llm #aiml #codegen #developers
We have just updated our PSX website https://phpsx.org/ with more examples and better documentation, if you are looking for a way to build type-safe APIs in PHP feel free to check it out #PHP #API #codegen #SDK
Is anyone using Apple’s OpenAPI code generation in an iOS app? I did some testing and compared it with other OpenAPI code generators for Swift, and honestly, it doesn’t feel quite right for an iOS app.
It’s very (very) powerful and solves issues like server-side generation and while the generated code is correct, it’s not very pleasant to work with. For example, compare @a_grebenyuk’s CreateAPI with Apple’s generated code, and you’ll see the difference.
Just released a new version of the polyglot data structure & bindings generator for hybrid #WebAssembly, #Zig & #TypeScript apps/interop. Now also supporting externally defined types for which only stubs for alignment & sizing are required, but which otherwise are opaque and can be used as any other type defs in this toolchain (e.g. embedded in structs/unions or as pointers, slices, arrays etc.)
Be careful when updating your swagger-codegen dependency, you might want to avoid
version 3.0.57 and stay with 3.0.56; fortunately my #unittests detected the problem
before breaking more, but it took me some time to find the cause and that an issue
has been opened already three weeks ago:
https://github.com/swagger-api/swagger-codegen-generators/issues/1295
The backdoor's source code ?
it was on GitHub
in a commit visible in a public repo
therefore
OpenAI might have been training ChatGPT on it *already*
or other folks training their own 'code gen' LLMs on it
"But I can just blindly trust whatever code snippet that this LLM recommends! Right? Right?!"
*cough*
So I have an interesting case where the Rust compiler decides to not inline a hardware intrinsic for aarch64.
This of course generates real sub-par code to do a call, but has other effects such as the code surrounding the call gets slower as well (setup of stack frame, store/loads to stack, etc)
https://godbolt.org/z/7sEna5TGv
The C version of the same code has no issues.
https://godbolt.org/z/4MPjcsf9z
Thoughts?
This week on The Changelog
Sourcegraph co-founder & CEO Quinn Slack on leading in the era of AI code intelligence 
This is a good one. Enjoy! https://changelog.fm/580
ChatGPT (et al) aren't great at Elixir, so we asked @josevalim about the possibility of building a custom "all knowing Elixir chat bot" for the community. His response 
Does anyone know of a tool that writes PHP classes out of diagrams?
I'm deprecating JsonSchema.Net.CodeGeneration. An attempt was made 
, but there's something better.
