81–90 of 152 posts
(and so no need to worry too much either way as the issue will sort itself out?)
Many have noted that LLMs are relatively good at writing code because they learned from all the code and other material, such as books, written before.
But let's say that we no longer produce the learning material, will LLMs be able to progress on their own or the slop they make (and read) will eventually lead to the decay of the models?
Will the models decay if left to their own devices?
I feel like the AI labs have an embrace-extend-extinguish model here like Microsoft of the 90s. Train your models on all the open source code, then extinguish the ecosystem since open source libraries are essentially a competitor. (If there were no libraries you'd have to ask the AI to do everything!) Like, personally I think the reason why AI works as well as it does for typescript is because it's essentially gluing…
> extinguish the ecosystem since open source libraries are essentially a competitor. (If there were no libraries you'd have to ask the AI to do everything!) That is neither the incentive of AI companies nor the truth. Availability of Open Source where stealing and illegal relicensing is not being litigated, is a perfect ecosystem for AI to work in. Maintainer exhaustion is totally a secondary effect, not intended. Th…
Malice, incompetence, etc. My question is, how much does it matter _why_ the problem exists because of AI/is being exacerbated by AI?
- They meet the localized needs of the original author or a certain subset of people. - They very likely will not be continuously maintained.
Earlier quoted context omitted.
I think you are correct on how it's playing out because AI cannot write software very well all on its own. The dream is that AI is so good that it can write all the code you need from scratch, replacing all and any code written by anyone else, but that's not happening
GPT-5.6 just finished writing a shim for me in rust that sits between Broadcom’s bullshit kernel and a standard Linux user space to turn my mesh routers into standard computers. At some point I’m not sure what the difference is in practice. Note: I can’t code. Not a line.
>"For years, the software engineering industry has operated on a comfortable, perhaps lazy, myth: that open source software is an infinite, self-renewing public good that costs nothing to consume and requires nothing to sustain." Since when? Open source projects have for decades offered paid support. Projects like Red Hat, Snort, Security Onion and others. I don't know anyone that has ever thought this. It's always b…
Yes, it's always been accepted that someone has to support it, but someone else, not me. But these are important dependencies so I'm sure someone will support them.
It's such a common situation that there's a six year old xkcd about it that felt about ten years late at the time.
I recently read Eric Raymond’s The Cathedral and the Bazaar. It was oddly sad to read. The enthusiasm and optimistic view of open source and the future of software and craftsmanship. Looking at it in 2026.. incredibly sad. Forget the bazaar. Back to the cathedral.
Nothing that was possible then is less possible or less potent now. The narrative is not friendly to communities of people owning complex software by sharing work now, but neither was it then. If you believe it was all wrong , an incorrect formulation, then disregard it and do not despair to move on. If you think ESR got something right that nobody can see anymore, then your hope should be rooted in the knowledge of…
So...the worse things are the better because we have more things to improve? Wouldn't it just be better if things weren't worse?
Can't entirely tell if this is supposed to be inspirational or contrarian (I think maybe the first one?), but every bit of shit in a situation being something we can look forward to overcoming by crawling through it isn't the most heart stirring way to look at it and also seems a bit dismissive without engaging with the actual subject.
Earlier quoted context omitted.
I think you are correct on how it's playing out because AI cannot write software very well all on its own. The dream is that AI is so good that it can write all the code you need from scratch, replacing all and any code written by anyone else, but that's not happening
GPT-5.6 just finished writing a shim for me in rust that sits between Broadcom’s bullshit kernel and a standard Linux user space to turn my mesh routers into standard computers. At some point I’m not sure what the difference is in practice. Note: I can’t code. Not a line.
Earlier quoted context omitted.
I think you are correct on how it's playing out because AI cannot write software very well all on its own. The dream is that AI is so good that it can write all the code you need from scratch, replacing all and any code written by anyone else, but that's not happening
GPT-5.6 just finished writing a shim for me in rust that sits between Broadcom’s bullshit kernel and a standard Linux user space to turn my mesh routers into standard computers. At some point I’m not sure what the difference is in practice. Note: I can’t code. Not a line.
Earlier quoted context omitted.
they are saying the prompt is the source, the code is the artifact. I think it is a valid comparison. when reviewing a PR, I do want to see the prompt(s) that generated the change.
My latest feature was an overarching change across 4 different stacks, about a full work day of going back and forth iterating on different designs, generating workarounds for stack/language specific quirks and the business' usage patterns. Multiple fresh agent sessions, review passes, exploratory work with subagents, etc. "The prompt(s) that generated the change" in that context is multiple hours of discussion, Q&A,…
The value is in how you came to the design you expressed.