Notices by Mallory's Musings & Mischief (malcircuit@thingy.social)

My opposition to LLM "AI" systems is fundamentally a really simple question:

*If the machine is going to do everything I value for me, what the fuck is the point of being alive?*

I haven't heard an answer to this yet that's either satisfactory to me or seems likely to actually happen.

I hate how so many people are struggling right now, while a handful of people have more money than it's possible to spend in a hundred lifetimes

I got a label maker and had the strangest experience. I opened the box, loaded the sample roll of label ribbon, inserted 6(!) AAA batteries, and it worked.

No apps to download. No account to log into. No firmware updates. No pop-over "helpful tips" about how it's "now enabled with AI" or whatever.

I just typed in the label I wanted and hit "print", and it did it. Astonishing. Nostalgic.

It is interesting and a bit strange that there is a very deep mathematical connection between how a stringed instrument works, how a radio antenna works, and how an LED works.

An LED is just plucking single electrons like a string of a guitar. Instead of creating a pure tone of a music note, it makes a pure tone of visible light.

The first time you encounter quantum mechanics, it might seem very odd that the universe operates via discrete packets of energy. How can an electron know what an integer is?

But a guitar also knows what an integer is. A string of fixed length L can oscillate with a wavelength of 2L, L, 2L/3, L/2, ... , 2L/n, where n is an integer. It's all harmonics; oscillating modes.

Sometimes I get annoyed when I can only run one telescope rig on a clear night, but then I remind myself that this "one" telescope rig is an insanely awesome chunk of technology and glass that many people would kill to have.

I'm unbelievably lucky and grateful to be able to do #astrophotography at all, let alone with the equipment I have. If all I had for the rest of my life was this one rig, I would still be more capable than most.

I went out with my #astronomy club on Saturday night and ended up staying up literally all night doing #astrophotography. I'm still working on processing everything that I captured because I kind of went hog wild and set up 4 telescopes (if you count my Seestar S30).

I stayed up all night because I wanted to try to capture a pic of comet C/2025 R3 (PANSTARRS). I was only able to capture for like 20 minutes before the sunrise blew out all the images, but I did manage to get luminance and RGB.

How many studies do researchers need to do before the threat of LLMs is taken seriously? This technology *might* have some useful niche applications, but widespread deployment will be a disaster for humanity.

This shit is an existential hazard, and not in the way the AI companies love to talk about. It's not going to take over the world like Skynet, it's a cognitohazard that turns anyone that interacts with it into an idiot.

https://www.psychologytoday.com/us/blog/the-algorithmic-mind/202603/adults-lose-skills-to-ai-children-never-build-them

Thinking about LLM-based "AI" chatbots as cognitohazards is apt. They are deceptive and subversive in an extremely subtle but systematic way. They warp your perception and cognition to make you _feel_ as if you're more capable while simultaneously degrading your skills.

I feel like the vibe of the 2020s is:

The rich and powerful handing out shit sandwiches to everyone and then whining that no one is being grateful for their generosity

This is always the thing I think about with LLMs. By definition, they are the statistical average of every thing ever written. Using one only pushes you into being mediocre. They are homogenizing humanity; eliminating any variation in how people speak, write, and even think.

@ngaylinn https://tech.lgbt/@ngaylinn/116284172328690293

This is something I was thinking about when I learned that Cory Doctorow is an adamant user of Ollama, an open source LLM (whatever that means), to do spelling and grammar checking on his writing.

If I were a professional writer with an established voice, I wouldn't touch anything based on an LLM for fear that it would subtly erase that voice. So slowly you wouldn't even notice, your writing will be barely distinguishable from anyone else's.

All of the most important writing in history has been at least slightly difficult to read. Any truly novel idea is uncomfortable to a degree. It often requires stepping outside of the status quo in some way and challenging assumptions.

LLMs never challenge assumptions. They are the assumptions crystalized — freezing and anchoring cultural development to one moment in time.

Using one isn't the future. It's trapping you in the past.

This is one of the main aspects of my philosophical opposition to "generative AI" and large language models. I don't care how "useful" they might be. Making my life easier or more productive isn't a sufficient justification to submit myself to a system that fundamentally does not respect anyone's unique experience and perspective. It's a system that's biased to enforce cultural conformity and stagnation, rather than embracing diversity and evolution.

@thomasfuchs @RnDanger @jonne @peter

I think the point of it is more to identify parts of a sentence that, while being grammatically correct, are worded in a way that's "hard to read" or whatever. An LLM would be very good at suggesting alternative ways to say the same thing.

But as you have pointed out elsewhere, that's essentially what an editor does.

I'm not arguing it's a good use of the technology, just that it's such a trivial application that it's not really worth talking about.

@RnDanger @jonne @thomasfuchs @peter

I feel like it's important to keep in mind the context of the article. He's using a spellchecker LLM. It's not a chat bot. He's not asking it questions. He's not asking it to write for him. It's like criticizing someone for using a keyboard app with autocorrect.

I also have a deeper philosophical opposition to most uses of LLMs, but a spell checker is such a trivial application that I'm having a hard time thinking of a reason it's "bad".

@RnDanger @jonne @thomasfuchs @peter

Yeah, I know nothing of Ollama, so it's hard for me to take a position on it. At the same time though, previous generations of spell checking and grammar checking are also arguably "AI" technology, and trained on similar sorts of datasets. I'm not sure whether a spellchecker based on lower complexity neural nets and Markov chains is meaningfully different from one that's based on an LLM. Seems more like just a matter of scale.

@jonne @thomasfuchs

I'm not going to argue the matter as if I perfectly understand and agree with Cory's position. He can defend himself. I'm just not sure that article says what was implied by @peter's post. It seems like Cory understands exactly why people hate AI, and he's trying to make a distinction between the technology and the application of it.

@jonne @peter

Yeah, I was going to say... I'm not aware of Cory ever saying anything like that about AI. I'm pretty sure he knows exactly what's happening.