Low-resource open source developers pretty much screwed now

cracauer@

cracauer@

Developer
All this AI craze has a very unfortunate side effect.

For 30 years you could do open source development on a potato PC that some windows user put in the trash. If you really wanted to rebuild the godzillas (chromium, rust, electron) you could spend $1000 on a mainboard/CPU/RAM combo that made it reasonably swift. You were competitive. Somebody sitting inside Google with a ninja macho machine did not have a significant advantage over you.

With LLMs that changes. You either need a monthly subscription to an AI service (and the $20/month option doesn't reach very far), or you need hardware to run an LLM locally. You'd spend about $3500, which gets you either:
  • An NVidia 5090 (FreeBSD or Linux)
  • A Strix Halo AMD Ryzen 395 system with 128 GB (unclear whether FreeBSD will do a Vulkan-backed LLM on here). This purchase also helps with compile times
  • A Mac Studio with 128 GB (stuck with macOS). No longer sold by Apple, now you need a more expensive MacBook Pro

That is very much unaffordable for many volunteer developers, especially those who are still students. If your project is popular enough you can beg Anthropic for credits, but only a small minority gets those. Hopefully you have parents with some money who want to push your carreer.


And that doesn't begin to touch on the second problem here, which is that those $1000 mainboard/CPU/RAM combos are gone now, the RAM is no longer included. Thanks to AI companies driving up the prices of RAM, SSDs and HDDs. So right now you cannot afford a chromium master for non-LLM work either.
 
A friend (a CEO type who stopped being a programmer a long time ago) says he uses one AI to write code and another AI to critique this code and claims to get much better results. He is bootstrapping yet another company but this time he is writing all the code himself instead of hiring / contracting. No idea how well this actually works....
 
IMHO, this analysis lack something. If an OSS developer wants to use AI mainly for pair programming, DeepSeek-v4-flash is so cheap, that with $10 they can work for few months, especialy if used with Aider-chat, that is very good in using prefixed cached inputs. So, they can retain some advantages of the AI (automatize repetitive tasks), without some disadvantages (they are still in control of the code).

For vibe-coding, yes, money is required. But, does OSS requires (always) vibe-coding? In the same project, some features can be produced by programmers using costly LLM models, and other from more traditional OSS developers. In theory, there is no money involved, only passion, so every approach make sense.

For commercial software development, it can be a problem. But also in the first years of PC revolution, the hardware required constant upgrades, because it cannot support new versions of Windows. The advent of Linux, that was more constant as resource usage, and the advent of very performant dual and quad core CPU, stop these needs. So, it is not a completely new situation. From a certain point of view, it is the IT sector that was an outlier respect other technological works, because the cost of hardware was very low, and "decreasing".
 
bakul said:
A friend (a CEO type who stopped being a programmer a long time ago) says he uses one AI to write code and another AI to critique this code and claims to get much better results. He is bootstrapping yet another company but this time he is writing all the code himself instead of hiring / contracting. No idea how well this actually works....
Click to expand...
lol

"How is your company structured?"
"Oh, I'm just using the ``wendelman-recursive-AI approach``. yours?"
"I'm using the ``knowing what dahell you're doing`` approach."
 
I write only for my own amazement.
I won’t have AI on my machines nor do I use either the freebie or subscription models.

I have zero interest in AI.
If you need AI then YouTube is jam packed with AI bullshit.
 
  • Like
Reactions: mro
OpenFreeNet said:
IMHO, this analysis lack something. If an OSS developer wants to use AI mainly for pair programming, DeepSeek-v4-flash is so cheap, that with $10 they can work for few months, especialy if used with Aider-chat, that is very good in using prefixed cached inputs. So, they can retain some advantages of the AI (automatize repetitive tasks), without some disadvantages (they are still in control of the code).
Click to expand...

Yes, there are cheaper online models. I haven't used them so they escaped me in my post.

OpenFreeNet said:
For vibe-coding, yes, money is required. But, does OSS requires (always) vibe-coding? In the same project, some features can be produced by programmers using costly LLM models, and other from more traditional OSS developers. In theory, there is no money involved, only passion, so every approach make sense.
Click to expand...

OSS work doesn't need AI coding (outside the odd utility function). But debugging help, code review, making test cases are all areas where it helps being "competitive".
 
JohnK said:
Why cannot one just simply not use Ai?
Click to expand...
That's my question, if we're talking about a more bespoke software package, just slap something on there saying that there's no AI and then commit to not accept any patches that you can identify as being AI created and call it a day.

AI of the sort we've got now is just flat out not sustainable and the only reason why we're still seeing so much pressure to use AI in society is that AI companies are being allowed to illegally dump their services by charging vastly smaller amounts of money than what it costs to operate the service.

Really, there's too much group think and all this AI is really just pointing out how in some ways the standards for developers used to be higher. It used to be expected that you actually knew what you were doing and could do the research to understand the problem that you were hoping to solve rather than just lazily type it into a program to solve and then verify that the solution is acceptable. Having watched people use AIs, it's really best not to touch that stuff, it makes people stupid and ignorant rather quickly.
 
enjoy drowning in slop i guess, none of these things have been worth a single moment of our time and every piece of software that adopts them immediately gets worse. going to continue writing software like normal here
 
cracauer@ said:
Even if you ignore AI, you still don't escape the problem of acquiring enough RAM right now.
Click to expand...
Its like watching pathetic gamblers who can't stop.

Is there a 1-800 Number for these AI addicted junkies? Need instant dopamine hit. Talk to a computer..
I do not get it.

The only thing worse than talking to a know-it-all is talking to a computer that thinks it knows it all.
 
I think this thread really belongs in 'Off-Topic' at this point.

Getting technical help on "FreeBSD Development" is one thing. Injecting your own opinion while the main point is providing technical help - that's understandable. But when a thread is more about expressing an opinion, than about providing technical help, that's called social media.

Phishfry : I've seen some Forums members complain about how people they know IRL got addicted to talking to AI's, and become difficult to tolerate.
 
The problem is people think AI is a godsend it is not. Look at the sensational title of this post. Do you think open source developers are screwed now???
Like that ain't an opinion? Right in a technical section?
 
A lot of this is VRAM development catered to GPU processors. Everything else is unchanged for the most part at the CPU and RAM as I understand it.

You don’t need local boatloads of computation. Just offload it to the cloud for on demand GPU processing.
 
cracauer@ said:
Even if you ignore AI, you still don't escape the problem of acquiring enough RAM right now.
Click to expand...
If you don't use resource hog compilers it is not so bad. And you can still get used computers (This month I bought two 8 year old mini PCs, 16GB each, one even has a 2 Gbe ports, for $40 each). Not very useful for vibe coding with local LLMs but good enough for a lot of other things.

Of course the situation sucks but I bet the semiconductor companies will catch up (right about the time people find out AI is not all that great) and we will have a glut of memory! Let us visit this in 5 years.....
bgavin said:
AI lets the lazy and sloppy produce a result quickly.
Click to expand...
All new tech goes through this cycle. Socrates argued *against* writing as it would make people forgetful, make them lazy thinkers and even give them a pretension of wisdom while remaining ignorant! One generation's luxury is the next generation's necessity. Whether AI follows this pattern remains to be seen but chances are, it will!
 
cracauer@ said:
OSS work doesn't need AI coding (outside the odd utility function). But debugging help, code review, making test cases are all areas where it helps being "competitive".
Click to expand...
True. In my reply I described the situation of today, trying to balance some points of your initial post. But there will be so many changes also in the near future, that it is very hard to predict what will be the future of (OSS) software development.

For example, probably also the development UI will be affected because today they are rather technical: shells, build tools, etc. But if in future, thanks to AI, everyone with a business idea can vibe-code it, then there will be simplified dev environment targeting specific platforms (web, mobile, desktop, etc..). Obviously FreeBSD as an OS will be developed in the usual way, but probably the desktop environments will accept AI vibe-coded plugins, created using a simplified path.

But there are many changes also in the cost of the tools. Local models are becoming more useful, and they require less resources. There are plans to create cheap hardware chips with encoded a specific LLM model, and an upper layer with some customizations to it. As already said, there are some cloud hosted models that are very cheap, and rather capable. Only the frontier models are very costly, and some of them are incredible. The frontier will continue to use a premium price, as usual in every business. But it seems that the market has solutions for every budget. There are also free solutions on openrouter, and also Google offer free AI. So there is diversification.
 
When I held Micron stock it was $8 and going nowhere so I sold it.

Years later Micron went over $900 due to AI. Huge demand drove this price and scarcity, same as happened to ammunition components, especially primers.

Manufacturers ramped up and now there is a glut of primers again. Prices will never come back to previous levels but supply is available again.

I recently bought 32gb of DDR5 for my grandsons new build…. 10x higher than the old prices. Same for SSD all due to AI demand.
 
You must log in or register to reply here.
Back
Top