Technology

Some of the contradictions:

artifacts have built-in politics deriving from their structure. A famous example is the nuclear power plant: Due to the danger of these plants, their needs with regards to resources as well as security power plants imply a certain form of political arrangement based on having a strong security force/army and a way to force these facilities (and facilities to store the waste in) upon communities potentially against their will.

But then:

That does not mean that it is impossible to take certain technologies or artifacts and try to reframe them, change their meaning. In some way computers are one such example: They were first used by governments, banks and other corporations to reach their goals but where then taken and reframed to devices intended to support personal liberation

The computers used in banks were used differently from the personal computers that made their way into our homes. They had different protocols in place into how they could be used and connect to the network, having to log your work, not leaving personal files on it, archiving work files every X years etc. So we see that it is indeed not the technology itself that's problematic, but how it is used, and I suppose this is what separates marxists from vibes-based leftism. If someone is punching you and you punch them back, you might both be considered violent, but we see that this violence has a character: someone was attacking, and the other was defending themselves.

Their argument also doesn't explain why the USSR was interested in machine learning and AI (yes, neural networks are not new, they were being tested with vacuum tubes as far back as the 50s and the USSR was very big into it) and why China is making so many models. They ascribe a universal character to different system instead of taking them in their particular material context. Frankly there is very little material analysis at all in both pieces, because neither look at things in their totality and in the material world. Any analysis or critique of AI that ignores what's happening in wholly different states just won't produce anything actionable.

They further contradict the earlier argument here:

A search engine scrapes pages to build an “index” in order to let people find those pages. The scraping has value for the page and its owner as well because it leads to more people finding it and therefore connecting to the writer, journalist, musician, artist, etc. Search engines create connection.

AI scrapers do not guide people towards the original maker’s work. They extract it and reproduce it (often wrongly). “AI”‘s don’t point out to the web for you to find other’s work to relate to, they keep you in their loop and give you the answer cutting off any connection to the original sources.

While the technology of scraping is the same, the purpose and material effects of those two systems is massively different.

(emphasis mine). They agree that things develop differently. But as for the difference they try to make between a search engine and AI, they are starting to twist themselves into knots there to try and keep up their deeply-held beliefs. There is nothing that says a search engine has to be this way, or that an LLM has to be that way. Secondly, Google has been under fire for years for trying to get people to stay in their 'loop' and not leave the results page. There are LLM providers that focus on making them into a search engine.

And then contradict this anyway:

Technologies are embedded not only in their deployment but also in their creation, conceptualization. They carry the understanding of the world that their makers believe in and reproduce those. A bit like an LLM reproduces the texts it learned from: It might not always be a 100% identical replica but it’s structurally so similar that the differences are surface level.

So they're indirectly saying a search engine only has surface-level differences from an LLM, which cancels their entire point about how search engines are actually different enough on the surface to be good while LLMs are still bad. That part felt like they didn't have a problem with the stuff they grew up with because it was normalized, but couldn't articulate it and so it produces this contradiction.

Another contradiction:

And freedom is not the only value that we care about. Making everything “free” sounds cool but who pays for that freedom? Who pays for us having for example access to the freedom an open weight LLM brings? Our freedom as users rests on the exploitation of and violence against the people suffering the data centers, labeling the data for the training, the folks gathering the resources for NVIDIA to build chips. Freedom is not a zero-sum game but a lot of the freedoms that wealthy people in the right (which I am one of) enjoy stem from other people’s lack thereof.

And two paragraphs later:

But the argument against using LLMs is not about shopping and markets at all. My not using LLMs does not influence anything in that regard, Microsoft will just keep making the data center go BRRRRRRR.

Exactly, the technology will exist whether we use it or not, you can't put the toothpaste back in the tube as they say. So what's the solution? Pretend it doesn't exist? Berate people into compliance? It's like saying I don't want to use drones in my war because they're pretty scary and barbaric. Sure, but the adversary is using them and will be very happy that you are not using drones against them (see Hedgehog 25 military exercise in Estonia). It's self-defeatist to refuse to use something because it's the "tool of the enemy", which is the grown-up way of saying it has cooties. They talk about not building the nexus torment, but what if building some of the nexus torment allowed us to destroy it for good? We know this intuitively as marxists: there will not be socialism without capitalism first.

On top of which are several words that betray what the author really thinks, such as putting AI in quotes (because it's not actually "intelligent" you see), or calling it stochastic, or the inclusion of the "(often wrongly)" in a paragraph I quoted. That's the sort of keywords you see repeated all the time on twitter from people who briefly tried chatGPT in 2023 when it was underbaked, made up their mind about it then and never tried LLMs again or followed what has been happening since then.