Comment on Portabase has been accepted into the CNCF Landscape! (open-source data backup and restore tool)
neo2478@sh.itjust.works 8 hours agoI did not read the original comment you are responding to. It is however known that all LLMs were trained with stolen data. They scrapped basically all of the internet with no regard to IP rights, or licensing terms. So from that perspective, anyone using LLMs is benefitting from work stolen from others. That being said, it does not mean you personally stole somebody else’s code and copied verbatim into your project. You are however, supporting the companies that stole, and are benefitting from the stolen data.
I also work in the software industry, and seem to be one of the few hold outs not using LLMs. It’s an ethical boundary I am not willing to cross for multiple reasons. So, because others are using it, it does not make it ethical for someone to use it.
With all that, based on your description of how you are using it, however, you are not vibe coding slop and instead are using it in place if what traditionally a junior developer might work on, with human oversight at the end. I understand why you are doing it, as this an open source project with limited resources.
I would say this is the least worst way to use LLMs. But by using it, you are still supporting a deeply unethical industry.
curbstickle@anarchist.nexus 8 hours ago
This is incorrect, there are quite a few models trained entirely on public domain and freely shared data. I believe you mean corporate LLMs, such as Claude.
This is the problem I personally have with the anti-AI crowd (and I’m not pointing at you, I’m making a general statement BTW), the industry is the issue. The technology is not.
To me its the equivalent of blaming all farmers for what corporate megafarms do. Or blaming all game developers for what the major studios do.
The technology is just that, technology. It is neither good or bad, it simply exists. The nonsensical claims of a AGI being here because they tried to redefine the term, or marketing LLMs as a do-anything solution, while effectively controlling the ram and GPU markets through manipulation, or ramping up data centers that aren’t closed loop, or a whole variety of other things - that has nothing to do with the technology and everything to do with the corporations behind them.
Also you didn’t see that comment because nonsense like that on a properly tagged post is a rule 1 violation and explicitly not allowed. They can be grownups and filter based on the tag, its why its there.
neo2478@sh.itjust.works 7 hours ago
I agree with your take. The problem is absolutely corporate LLMs. I think the farming analogy is a good comparison. Though with all LLMs there is still a bit issue with hallucination, so its like a local farm that produces 30% poisoned food, for the case of the public domain models.
I have been trying to be more careful with my language around this topic. Because AI in general is very useful. There’s plenty of good coming from ML for instance.
curbstickle@anarchist.nexus 7 hours ago
Even hallucination thats predominantly a temperature setting issue, set low they will respond with “unable to answer” or similar. The big ones out there allow higher so it sounds more authoritative.
I build local models (for edge and device-specific purposes) with open datasets and my own, so I adjust the default temp accordingly.
That said, marketing has really pushed the idea that LLMs should be considered accurate, when really it should be considered a first draft at best. I don’t care if youre generating a summary or writing code (though I personally don’t think fully generated code is a good idea, it becomes a maintainability issue even if it “works” but portions are fine especially with a person reviewing).
Anyway, yeah the issue is corporations and market manipulation, as is tradition.