Yeah, “which of these seven rewrites of FastQC should I be using” is going to be fun

But surely you know all 21 of them, right?

Paul Hoskisson discusses the Streptomyces genus; the soil microbes behind many of our most important antibiotics and a cornerstone of the antibiotic revolution.

Excerpt from Matters Microbial 125: Antibiotics, Streptomyces and Evolution
@markowenmartin.bsky.social @paulhoskisson.bsky.social

Matters Microbial #125: Antibiotics, Streptomyces and Evolution YouTube video by MicrobeTV

A new episode of my #MattersMicrobial podcast is available! This time, Dr. Paul Hoskisson joins the #QualityQuorum to discuss Streptomyces, antibiotics, and evolution. Please spread the #GoodMicrobialWord? I truly can use views and listens to keep podcasting. Thank you.

youtu.be/2w9Hll4xcFk?...

Well played, sir!

I've spent a significant chunk of my career in scientific software development trying to convince PIs that the code their students/post-docs write should be considered a prototype to prove the concept. And prototypes are designed to be crashed and thrown away. My success was rather limited.

Sure, but the thread mostly is about using LLMs to rewrite someone else's code. For your own code, you'll know best what makes sense. I've yet to get a patch out of a model that I'd be happy to merge for anything non-trivial.

Is this a "guns don't kill people, people kill people" argument? 😉
I certainly do see more PRs without seeing more mergeable PRs.
Anyway, I do like your attempt at defining some guidance. I just think that for most of my projects, my guidance would be "please don't vibe-rewrite this"

Vibe coding clean-up specialist?

My experience in projects that I'm involved in is that people do feel empowered to submit changes. Unfortunately, their AI tools often can not produce code that can be merged and maintained. This leads to conflict and an even higher burden on maintainers.

The practices you propose are very reasonable. I just have limited confidence that people will live up to them. Validation is hard, and "drop-in replacement" is a big claim. I'm saying that as somebody who has indulged in a bit of "let's rewrite it in Rust" myself.

Now, I'm not saying that this setup can't work for something like RustQC. Such a tool might be easier to validate even if output isn't byte-for-byte identical. But that's not true for every bioinformatics tool out there.

I think you also used the term "nearly identical" somewhere in your article or thread. As someone who builds complex tools, the thought of someone pitching a rewrite of my code with "the same output, kind of" is not very comforting.

I'm not looking forward to a future where all the tools are being vibe-rewritten into languages people don't want to learn. Who will maintain all this? The original maintainers won't. It's not the language they were comfortable with. Does the prompter understand the tool well enough?

Whatever happened to “use the right tool for the right job”? F1 cars might be great for racing, but I wouldn’t want to take one grocery shopping.

📢Come and join us in Mexico later this year for the 5th Series of #SBNP!
Registration is now open - secure your spot today!
👉 Click here to head to the conference website for more information: bit.ly/4v25JYK

As a benefit, everybody in this process deserves what’s coming for them, and it’s a code base nobody will miss. Do either way, it should be fun to watch from afar.

I’m kind of interested to see what happens with that project that vibe-transcoded the leaked Claude code TS code to Python to circumvent Antropic’s DMCA takedowns. Antropic really seems to hate that leak, but going after the vibe-transcode would set a precedent for others to go after their customers

I feel like I was told that you needed to invent a file format, write a parser, and create an ontology to be a proper bioinformatician before that company was even founded...

Every time I hear "data driven", this gif pops up in my head.

And people ask me why I don't enjoy web dev.

I don’t think I want to click through, but this is a reaction to that sexy taming of Gollum painting that is making its rounds right now, I hope.
I don’t think I can handle another front in this sexy LOTR timeline

Also I hate how that’s my first thought now when people I don’t know send PRs

That’s how supply chain stacks start 😉

But the mode of action seems to be via interfering with the shikimic acid pathway. Not sure how that would give cross-resistance with many clinical antibiotics. A quick pubmed search mainly gave a Raoult paper, not sure I want to read that.

So your Mozart number probably is 3?

Someone needed a tech front for their illegal organ trading business.

Wait, there are people who don't use it as a slur?

The discussion threads full of "but I need Claude to finish my homework" are funny, though. Especially the ChatGPT-written longer ones.

I explicitly tell my students that they won't learn how to program if they let the AI do it for them.

It's crazy how good it looks for 10x6 pixels.