a-tarasyuk wrote: > Was an AI used at all in this PR?
@ojhunt Since I’m relatively new to this project, I’m not exactly sure what level of detail is expected, so I’ll give both a short and a detailed answer. _Short:_ No, AI was not used to implement this PR. Do I use AI tools in general? Yes, I do. _Long:_ I’ve been contributing to compiler projects for about eight years. During that time, I’ve made many patches to the TypeScript compiler, including both the [original version](https://github.com/microsoft/typescript/pulls?q=is%3Apr+author%3Aa-tarasyuk+is%3Aclosed+is%3Amerged) and the [new port](https://github.com/microsoft/typescript-go/pulls?q=is%3Apr+author%3Aa-tarasyuk+is%3Aclosed+is%3Amerged). The official [release notes](https://devblogs.microsoft.com/search?query=a-tarasyuk) highlight many of the significant changes delivered during that time. I do all of this in my free time while working full-time for a company unrelated to these contributions. For the past year and more, I’ve also been contributing to Clang to broaden my expertise in different programming languages. To streamline my workflow and focus more on the technical aspects, I’ve developed my own work patterns that help speed up the process from initial investigation to opening a PR. For example, in many cases, I use a structured PR description template and prepare detailed notes during debugging. This allows me to finalize the PR description upfront, rather than revisiting it later to add missing details, as some of my earlier PRs occasionally lacked that information. It also helps reviewers by providing clear context from the start, reducing unnecessary back-and-forth, and maintaining a clean and consistent commit history. Before publishing a PR, I use an AI tool called [Grammarly](https://www.grammarly.com/) to check grammar. I may also use AI to quickly find relevant specification history with direct links, which greatly speeds up research in large specifications documents. Have I tried using AI to fix real compiler/dev tools issues? Yes, I have. But mostly, it doesn't work well. AI often offers surface-level fixes without finding the real cause. Right now, it doesn't speed up my workflow - in fact, it can slow it down, especially when the model lacks enough context and keeps suggesting generic or unrelated solutions that don't solve the actual issue. I hope this provides the necessary details. Please feel free to let me know if any further clarification is needed. https://github.com/llvm/llvm-project/pull/182362 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
