@thomasfuchs
But then AI companies came and destroyed all of that. Apparently copyright law doesn't exist if you operate piracy at a large enough scale. I think there are two parts to why "we" accepted it. First was "Open" AI's weird corporate structure with the 501c3 that clouded the for profit motives. Second that the entire planet is being held hostage in a software cold war where the, legitimate, concerns are that rival countries will have more powerful AI that can exploit vulnerabilities
15
Replies
1
Boost
@thomasfuchs I do think free, open source, and legal (or ethical if you will) LLMs, training data, and models would change what LLMs do.
All of the commercial ones have two issues, the unethical side (you know the drill) and the socioeconomic impact. Ie it’s meant to replace people and tailor what data we have access to (since it can server up whatever it wants).
How would LLMs work (as in what we do with them) if it wasn’t driven by greed and propaganda? What would be done with it then? Would it be done differently?
Not sitting on answers, but it’s a question I think we should talk about more. The lock in was why open source came to be after all.
@thomasfuchs it may come as a surprise that my core opposition to LLMs is not at all based on environmental/economic/etc destruction but on them being fundamentally incapable of connecting to reality: the best an LLM can ever hope to achieve is sounding *plausible*, including "corrections" ("sorry for making that mistake, it won't happen again" - to borrow a term, this sentence is *not even wrong*). IMO there is literally zero use of this tech that's not detrimental to users' minds.
@thomasfuchs I've written a bunch on this -- for example https://www.quippd.com/writing/2026/04/08/ai-code-is-hollowing-out-open-source-and-maintainers-are-looking-the-other-way.html - but a bunch are linked from the main site if you remain curious.
@thomasfuchs
My biggest frustration about the AI situation, aside from people offloading critical thought to it, is copyright has been completely ignored. For years I've been learning minutia of how US copyright law works for preservation. I only publish materials through archive.org because they're a registered 501c3. I've looked into what it would take to run my own legal archive as well. For videos I've also read more into what properly counts as "fair use" and try to work within those laws
@thomasfuchs
But then AI companies came and destroyed all of that. Apparently copyright law doesn't exist if you operate piracy at a large enough scale. I think there are two parts to why "we" accepted it. First was "Open" AI's weird corporate structure with the 501c3 that clouded the for profit motives. Second that the entire planet is being held hostage in a software cold war where the, legitimate, concerns are that rival countries will have more powerful AI that can exploit vulnerabilities
@thomasfuchs
It is indisputable that AI companies are training on copyrighted materials, and that they seek to profit from it. But they have been deemed untouchable for this crime which has only emboldened them and investors into thinking they are unstoppable. The only solution I see is to nationalize these companies and make access free. But "muh freedums" will call that "socialism" while the entire economy, from fab throughput to electricity usage, is pillaged for infinite corporate growth.
@thomasfuchs
PS: To pre-counter "but humans train on copyrighted material". Humans can be sued if they directly copy their training material
- The street vendor selling air brushed Elsa shirts is obviously breaking the law
- Parody laws exist as a carve out to protect works that transform existing material
- Clean Room engineering exists because just seeing the source material "taints" you
The law has been VERY clear on this for humans. But automate it at a huge scale and suddenly "it's okay"
@thomasfuchs > "it's not ethical to..."
...waste time fine tuning some prompt/writing pages of dialog to get the damn bot to extrude, e.g., the code you need while writing those description in a language, English, which is imprecise and poorly suited to describe the problem at hand (in my example: compared to programing language), when you could be spending that time slowly explaining it to a junior human instead, thus transferring know-how and helping creating a new future expert in the field.
@thomasfuchs I wrote about different ethical dimensions at play in LLMs a little while ago, sparked by a conversation on here:
@thomasfuchs this nerd-snipped me into a blog post. it's not hard line ethical reasoning. more taking a moment to re-asses where things are and how I'm interacting with them.
@thomasfuchs The hard thing about discussing ethics is that there are so many ethical systems to choose from. America used to feel like a mix of Utilitarian and Deontological ethics, but there's a growing faction of Virtue Ethics fiends, the ones who think "we're good people, so anything we do is good, and nothing we do is bad."
I find it best to think of the LLM people as predators, and we're their chosen prey. Or as they say in Ukraine, "the people are the shit the oligarchs grow their money in". There is no ethical system that can convince me that it is right and good for a predator to devour me. That's slave bible territory. You don't discuss which rights you deserve with Nazis – you punch them in the nose until they go away.
We can talk about IP theft and ecological destruction all day, but that feels like arguing about the ethics of what kind of weapon a mugger should use while beating me up. LLM-AI is a scam. It was built to be a scam. It doesn't matter if some of the pieces are ethically acceptable in isolation, because the whole system was built on an unethical foundation with unethical goals.
@thomasfuchs I think for me the main issues are environmental destruction, and what for want of better terms I will call plagiarism. I'm not interested in "per-inference" costs as long as enormous, energy and water hungry data centers are being built or proposed. Even solar and wind power (although vastly superior to fossil fuels and nuclear power in this sense) have significant impact at scale. I'm also skeptical about "local models will save us" on several grounds, unless the modestly powered machine ingests only data whose creators specifically consent to that use. Yes, it's impractical, but I thought we on fedi already had the "consent is impractical" discussion a few times. Finally (and you did invite unstructured!), I don't want to directly or indirectly support the businesses of some of the worst people on the planet. Every hour I spend reviewing AI generated PRs makes these services a little more useful. I'm not ashamed to say that I just want to the AI bubble to pop as goal, not as an means to some other thing. We didn't boycott South African wine because it tasted bad, but because it enriched terrible people.
P.S. This post is not an invitation to "Change my mind", but a response to a request for thoughts about AI. Reply guys will be blocked.
@bremner all good points! thank you
@thomasfuchs I think the problem here is separating the technology from the organisations that control them.
I don't see anything intrinsically unethical about a given llm model or any machine learning model. It's code.
The ethical issues arise from the motivations, attitudes & goals of those who wield the technology. Those who set the design parameters of the models, & who look to exploit them.
Right now AI looks to be largely a fraudulent financial engineering exercise. Sub prime tech.
@thomasfuchs A thing that has started to come up a lot in my workday is that these tools become corrosive not just to the quality of the production output, but to the relationships between people. With people that I know have enthusiastically embraced these tools in my newsroom, we're always fighting against distrust that we're not actually talking to them, we're getting orders and requests from the bot. That tends to build resentment, it's not a good environment for collaborating on either software or journalism.
