Why Open Source Maintainers Thrive in the LLM Era
mikemcquaid.com
mikemcquaid.com
Interested if any other maintainers have a similar experience?
Daniel Stenberg definitely hasn’t.
https://www.linkedin.com/posts/danielstenberg_hackerone-curl...
This is a contender for the most disappointing writing I’ve ever seen from Mike (which isn’t a regular occurrence). It completely misses so many important factors which have been discussed ad nauseam, such that someone abusing LLM tooling today—especially a junior—is crippling their own learning. But all those arguments pale on the face of this blatant embrace for profit above all. I’m profoundly saddened these are the views of someone who is at the helm of one of the most popular open-source projects currently.
The whole article lacks any valuable insight and reeks of the “my business uses AI” hype so many companies are chasing just to be valued and get attention. This is not about “open source maintainers” in general, as the title suggests, but about Mike’s personal experience.
> Let’s build some cool shit (and faster than we could in 2020).
No, let’s not. Let’s go build some stable shit for once. Everything is broken, and you’re partying like breaking everything some more is a good thing.
LLMs are a tool. They can help someone drive a nail through a piece of wood or their own hand. You can use them right or wrong, effectively or ineffectively. But you’re fawning over them like it’s all a panacea and completely ignoring how many people are proudly and ignorantly using them wrong. One day, not too far now, one of those people is going to drive a nail through your hand.
I don't agree with this if they follow the guidelines I've discussed in this post about e.g. actually reviewing and ensuring they understand the output of the LLMs.
> This is not about “open source maintainers” in general, as the title suggests, but about Mike’s personal experience.
You'll be unsurprised to hear I gently disagree here. It's not based just on my experience but the (very mixed) experiences of my peers. Those who are good at code review do seem to be having a better time with LLMs.
> Everything is broken, and you’re partying like breaking everything some more is a good thing.
My experience with LLMs has been that they help me fix broken things more quickly than before LLMs. Again, as I mention in the post, if you're not reviewing the output here: you're doing it wrong.
> You can use them right or wrong, effectively or ineffectively.
Exactly. This post tries to explain how to use them effectively, something I find OSS maintainers find easier.
> [You are] completely ignoring how many people are proudly and ignorantly using them wrong.
The point is precisely that too many people are never going to verify outputs and will even resist any kind of human review to their LLM-generated code. This is not theoretical, we know this is happening. Which would be fine in a “you do you” manner if what those people did only affected them, but it affects everyone else too. Because we don’t write every software we use, and some day soon we’ll be bitten by one of these idiots who introduced a major security flaw in some system we’re forced to use (e.g. government website).
In other words, what I’m objecting to is precisely the narrow view of this article in unambiguously propping up the good parts and being blind to the bad parts, even criticising those who have a concern for the bad.