Hello & invitation :)

[dsyme]

Hi @Pencilcaseman

I know I've thrown a few AI reported bugs at the repo, so I thought I'd introduce myself and explain what I'm working on, and ask if you'd like me to continue the work or to collaborate together.

I'm Don Syme, a researcher at GitHub Next, a research group in GitHub. Our group is working on the topic of Continuous AI, with a current focus on one particular AI coding technology called GitHub Agentic Workflows, the repo is here and our sample pack of demonstrator workflows is here.

Effectively these are running Claude Code or other agentic processors with prompts in GitHub Actions.

The two workflows I'm currently focusing on are

• Daily Test Coverage Improver
• Daily Perf Improver

Both are aimed at helping repo maintainers achieve higher quality, in a way that is under their control, without necessarily adopting AI/LLMs/Agents for their main coding.

I came here because I chose a C++ math library at random to try things out on, simulating the maintainers role in the fork https://github.com/dsyme/librapid. Hence the PRs, Issues in that repo.

I'd like to ask you if you'd like me to continue with this work, or if "enough is enough" :)

• I can wind up the work if you like, as is, or
• I can get the test coverage improvement PRs moved over to this repo for your perusal, or
• I can get the simpler bug reports and PRs moved over, or
• Or possibly we could take this all further and look at actually pushing the perf improvements identified into the library. This is a more serious thing as it needs careful review and would have to actually align with your goals for the library

Please let me know what you think

Thanks again for letting us use the power of OSS to use this as a test case.

Best
Don

[Pencilcaseman]

Hi Don,

Sorry for the late reply -- work has been pretty busy :/

I haven't worked on LibRapid properly for quite a while now, and, as much as I hate to say it, I doubt further work will be of much benefit to the OSS community.

That said, it's probably quite a good testing ground for the sorts of things you're working on -- I suspect there are an awful lot of bugs I've introduced by accident (I'll have a look at the issues you've opened when I have time) and there are probably some nice little performance improvement opportunities in there as well.

I'm more than happy to keep working on this with you if you're keen, but you may have to put up with some slow replies (at least until my internship ends.)

Let me know what you think :)

[dsyme]

@Pencilcaseman Thanks for replying!

That makes total sense - I understand how these things go.

It is an interesting use case and I'm wondering if it could be an interesting case study for a fully-automated AI spec, test, perf experiment. These things cost money and time but the results were interesting and encouraging. However some thought is needed about the automatic AI dev infrastructure - we'd really need access to GPU and a variety of CPU spec machines with different capabilities, or to settle on one machine capability.

FWIW I conducted some loose automated experiments in language implementation and design - basically AI-automating the production of a programming language. Like this it's at the borderline of what can be done today.

What are your plans for after your internship? Maybe we could chat on teams?

[Pencilcaseman]

Yes I imagine the cost for these things isn't cheap; especially for R&D.

I don't currently have access to a proper GPU, so that might be a bit tough.

After my internship I'm back off to university, unfortunately, so I'll once again be a bit short on time (though less so than currently.)

I don't want to commit to anything I can't uphold, so it may be best not to focus too much on LibRapid for the time being if you think there's a lot of work involved.

[dsyme]

@Pencilcaseman OK, sure! Good luck with your studies, I hope our paths cross again!