Comment by tpmoney | Hacker News Reader

Comment by tpmoney | Hacker News Reader

Parent Full thread

And often the most useful software is closed source proprietary things that cost money to be used effectively. That hasn't stopped open source from building impressive and useable systems that are serviceable enough for most use cases. You can't fork photoshop, but GIMP (for all its warts) exists and you can use it for photo manipulation. You can't fork Solidworks, but FreeCAD and OpenSCAD exist, and you can use them to do 3d modeling. You can't fork Eagle, but KiCad exists and you can design your own PCBs with it. You can't fork Windows or MacOS but Linux and BSD exist and they probably already run most of your hardware. You can't fork Oracle, but Postgres and MariaDB exist and you can run a business on them. Unless we stupidly legally lock AI training and model creation to massive proprietary corporations, why do you think the same won't hold true for open AI models and projects? Maybe Claude will always be a better code agent than AInix or whatever, but as long as AInix is good enough, that doesn't matter any more than it matters that Solidworks is better than FreeCAD.

I see little real difference between using proprietary software and using AI to generate software in terms of the freedoms allowed to the end user. The entire point of the free software ethos is that the end user should be in control to the degree that they want to be, not merely to the degree that a proprietary service allows. If nothing else the obliteration of programming culture and education and the mass deskilling created by the normalization of AI works against that ethos. It turns programming into just another form of capitalist consumption.

Which is all well and good if you don't care about such things and just want the end product but AI shouldn't be conflated with FOSS. A tool that strips knowledge away from people is antithetical to the culture even if the end product is high quality.

AI is a tool like any other. AI doesn't inherently "strip knowledge away from people" any more than an IDE, or a GUI or a modern computer does. Yes, you can use AI to blindly do things without learning. And you can use GParted to blindly format a drive without ever learning what cylinders and sectors are. And you can use GIMP plugins to blur images without ever having to learn matrix multiplication. Or you can use drop box and never have to learn how to use rsync, samba and chron. But you can also use AI to learn things like which obscure emacs functions you can string together to customize your emacs config in a way that suits you better. Or how to programmatically invoke disabled junit tests from within another test so that you can test that an annotation that decorates a unit test does the things around that test execution that you expect. That last by the way is a real thing I did, and something that is difficult if not impossible to search for with normal search tools because the terms that you need to use are much more likely to get you results for generally enabling and disabling unit tests.

If a user chooses to use AI in a way that strips knowledge away from them, and doesn't contribute to their learning, that is a choice they are making but there is nothing inherent in AI that requires it anymore than choosing to code in a higher level language inherently prevents you from learning assembly language.

Reply on news.ycombinator.com