Post

I ported a Python library implementing a full HTML5 parser to JavaScript using GPT-5.2 and Codex CLI in 4.5 hours, and decorated for Christmas and watched Knives Out while I was doing it simonwillison.net/2025/Dec/15/...

I'm sharing this because it's quite a stark demonstration of what's possible using the latest frontier LLMs at the tail end of 2025 Is it a GOOD idea to churn out a 9,000 line direct port while watching Daniel Craig chew on the scenery? That's a whole other set of questions!

I guess we're calling this "vibespiling" now! simonwillison.net/2025/Dec/17/...

... and now we're up to four implementations: Python (Emil Stenström), JavaScript (me), OCaml (Anil Madhavapeddy) and now Swift as well (Kyle Howells) simonwillison.net/2025/Dec/18/...

This raises a funny question, what if someone tricks AIs to copy an entire library line by line, would they own it? This could be done to say, bypass the GPLv3 license.

One argument I've seen against this kind of contribution is that people start making PRs full of slop content that doesn't work, but it takes time to determine it doesn't work, and the added effort this requires from maintainers slows down the acceptance of actual useful contributions

I really like the term "extractive contributions" for that - "those where the marginal cost of reviewing and merging that contribution is greater than the marginal benefit to the project’s producers" simonwillison.net/2025/Oct/2/n...

(IANAL) Early rulings would suggest the copyright part is in a *truly* wild new bit of territory, in that the initial rulings around art stuff (IIRC including the one with Anthropic last year) landed on the work being transformative, fair use… and *uncopyrightable*. We have some figuring out to do.

Have you happened to read this post about the various JustHTML ports? I’m curious if that changes your thinking about 9,200 tests providing a sense of safety. I write a ton of AI-generated code and I always worry that a test suite full of tests I don’t understand may come back to bite me.

I was never planning on using a mainly unreviewed vibe coded HTML parser for anything that mattered, but the differences described there don't seem to be particular surprising or critically important to me

Simon, have you seen this article? When a person asks LLM to port a project to another language, they get no guarantees that it doesn't hallucinate some parts and it doesn't break other behavior not covered by tests.