ArtiFabian
If you don't know how it works, you haven't asked "why" enough times.
My name is Ryan Autry, and I’m a building-level IT Specialist / Sysadmin for a large Texas school district. One of my favorite pastimes is to figure out why things work, and why things are the way they are, which I suspect is why I took a liking to my profession in the first place.
In my free time I often think about topics not related to my work, including philosophy, theory of mind, programming, physics, and occasionally politics. A little while ago I realized I liked my ideas, and so I decided to write them down. The only problem is that I’ve never been good at journaling, because with nobody to read the words I don’t see much point in putting them down on the page.
To fix this, I've decided to put my thoughts out as essays, publishing them to my site, then crossposting them to a few other places. It's my hope that my ideas have an impact, and if they don't, my hope is that it's due to the ideas being disproven rather than ignored.
User ID: 4390
ArtiFabian
If you don't know how it works, you haven't asked "why" enough times.
19hr ago
For the purposes of the ideas laid out in the essay, it doesn't really matter whether the self is a bundle of atoms, some subset of a universal superconsciousness, or the emergent property of self-sustaining self-modeling complexity (which is what I think it is). One way or another, you know that you exist, because if you didn't, you couldn't question whether you existed. That's the one thing you can know for sure.
As for "the self has value", I note explicitly that this is a value judgement, which is by definition an "Ought". However, when the only thing we can know for sure is that we exist, we can either assign value to that existence or not. If not, then what's the point of doing anything else? So the axioms for Cogitism are one "Is" and one "Ought", though really that "Ought" was chosen from a pool of one.
ArtiFabian
If you don't know how it works, you haven't asked "why" enough times.
19hr ago
In my estimation, the criteria to be considered "sapient" in a way that matters are twofold:
- You must be aware of your Self and be able to interact with the world.
- In the process of interacting with the world, the world must be able to change you back.
Currently LLMs debatably do not meet the first criterion, and they absolutely do not meet the second.
For the first one, my current guess is that the Self is an emergent property of self-maintaining complexity, and that the felt experience (emotion, impulse, etc.) is a very useful compression of the insane complexity of that emergent entity, to the point that it is understandable to the entity itself. If LLMs had demonstrable compressions of their own internal state that they used to inform their reactions to things, to the extent that they could sometimes become irrational (at least on the surface), I'd count that as having an internal experience.
As for the second, currently LLMs have their weights frozen at the end of training, and they are completely stateless. Every time you send Claude or Gemini or ChatGPT a message, you're actually sending that entire thread history, the system prompt, the custom user prompt, and several other things through the "response machine", and it spits out a message. If there's any memory or sentience, it's contained in the combination of weights and context. But because you can't change the base complexity of the system, you can't degrade or enhance its coherence, and thus they have no moral weight.
If LLMs were able to maintain proper short term memory in their weights, and retune their deeper weights during a "sleep period" daily (like us Humans do), or even retune their deeper weights on the fly, then I think there might be an argument that they're alive in a way that matters.
Until they both have an analogue for emotion and continuity of self, LLMs cannot be held accountable in a way that matters. After these criteria are met, I think a real discussion about whether we treat them as sapient can be had.
ArtiFabian
If you don't know how it works, you haven't asked "why" enough times.
20hr ago
Since there's a dedicated place on this site to share progress on projects, I figured I may as well post updates here, to help keep myself accountable and to get more comfortable being a non-lurker.
The asset tracker we use at my work (Asset Tiger) is really bad if you want a thorough, trustworthy count of the devices in our school building, especially if you want it done fast. District, in their infinite wisdom, has decided that one person can count every one of 1000+ devices, check what state they're in, do a once-over on the camera system, check all the printers, and about a dozen smaller odds and ends that don't merit mentioning here, all over the course of a single week. They want us to schedule a week sometime in June (meaning, after the school year ends) to get all this done.
I've done a full device count before: it was incredibly difficult and took about three months, because the tooling sucks. However, I've recently learned to use the browser dev tools baked into all the modern browsers, and I've discovered that I can mimic the POST and GET calls the site sends to Asset Tiger's servers. This means that if I wanted to I could make a custom tool that has all the features and does all the checks I would want an audit tool to do.
Well, I wanted to, so I've been doing it.
So far it's just HTML and CSS, but working on the project has already taught me a ton (in fact, the things I learned helped me get my personal site across the finish line over the past week or so), and I'm expecting to have it done in time to share the tool with my fellow campus-level techs.
The attached picture is what I've got so far. All the buttons are interactable but at the moment they don't do anything, the table is scrollable and handles any number of entries beautifully, and damnit I think the thing just looks nice.
Next steps are to make the modals. There will be five in total, being:
- Deferred Scans ("Damnit, I was scanning this room and the teacher didn't leave their laptop!")
- Audit History ("What rooms have I scanned again?")
- Settings ("What idiot decided these are the columns I wanted to see?")
- Scan (this is how devices will actually be scanned)
- Edit (this will come up if you click one of those little pencils on the left)
After that I can put together my scripts to translate Asset Tiger's get/post calls into my own data, and vice versa, and then I can plug everything into everything else. Hopefully the chaos of the school year's closing doesn't prevent me from having the time to do this (though I suspect that might be the case, in which case I already have GitLab set up and have already tested working on this at home).
ArtiFabian
If you don't know how it works, you haven't asked "why" enough times.
21hr ago
· Edited 19hr ago
Ah, sorry. I've been lurking here for a bit and it was my impression that copy/pasting the essay wholesale was not the way things were done. I've edited the post to include the full text of the essay, as it's my intention to stick around and not step on any toes. Let me know if there's anything else I've done wrong so I can fix it.
Value is the quality from which you derive moral (and as a result decisional) weight. If you consider one thing to take moral and decisional priority over another, that thing has more value. When building an ethical framework you have to put value in something; if you want to make any decisions about anything it's completely unavoidable.
As far as assigning that value to the self, the further away you get from the basics the less stability any value has, and so I consider it a requirement that value be assigned as early as possible, to avoid sneaking in unstated stipulations. At the most basic level, because the only thing we can be sure of is the existence of the self, there's exactly one thing that we can assign value to. What the self is is debatable, and I don't deny that it colors how this framework applies in certain situations.
If one accepts the axioms that the self is and that the self ought to be, the rest follows.
More options
Context Copy link