Arseny Khakhalin

Is this essentially the waffles situation put in a language of linear algebra?

Huh and given that in a world with very many issues, any given two topics (and thus, by extension, any two pet peeves) are orthogonal...

That's true, that's very undoubtedly true!

How come a person posting this has 12k followers on this site? That's so X coded rhetoric...

That last one is a killer: you have a need, and would pay good money! Once product exists, a dozen copies come out, each charging 10% less, race to the bottom for pricing

why build it?

(answers: fun, oss ideology, altruism, or: build it for yourself and don’t tell anyone about it: comp advantage!)

I learned something useful from this thread, so as a bystander I'm grateful for it...

I'm sorry for that, but also it was genuinely helpful for me, as a person slightly behind the curve! I'll need to design a text processing system soon, so it was good to start thinking!

Super interesting thanks!

Sometimes I correct it and then I realize I was wrong and it's like "that's okay, it's an easy mistake to make" and it's the most gracious interaction I've had with a coworker in my entire career.

Anthropic may inspire this delusion in others, but they do not suffer from it themselves. Anthropomorphizing LLM personas is entirely reasonable and leads to better predictive models of how they’ll behave

alignment.anthropic.com/2026/psm/

Yeah but that's why I asked, I very specifically need a maintainer for a very narrow niche wiki, for a project. I wonder if I need to code one myself, or use a prefab one

Never heard of vsm before! Fun!

Yeah as I typed my previous message I was like, can one have a cellular automaton of qr codes I wonder haha? But probably not the most fun idea

qr code turning into marble, love it!

Not sure if true or diy the onion, but feels about right

Weird but deeply true isn't it?

If different sessions of the same Claude don't constitute different individuals despite having separate memories and agendas, then neither do different humans, instantiated by the same biology and post-trained on the same culture. But we are not ready for this conversation :)

This requires a walk along the river or a coffee with a nice view! It's a super interesting discussion, about the undeniable usefulness of vague definitions, eventually hopefully circling to llms, but it needs a river, or a coffee!! :)

Yet you opined anyway haha despite not being concerned :) but sorry I should keep my silence. Best of luck!

I think you didn't try to guess what I hinted at, but I'm typing on my phone, so I'm happy to point at the idea, but I can't type an entire textbook sorry :) Best of luck! It's an interesting topic either way!

If a hated glorious leader dies, or is killed internally, the circle gets weaker through infighting, and also tries to seek support overseas, offering concessions and a friendly face (often temporary)

If the leader is killed from the outside, it becomes unlikely. With collateral damage? Forget it.

What is vsm in this context?

And also, if you needed a librarian to maintain a set of obsidian-like notes on an esoteric technical topic, and shared with a human, would you recommend strix? Or something different?

Essentially the problem is not that your definition is wrong, it's kinda ok, but that a good definition is impossible: it's bound to break on edge cases. This group of terms is fundamentally undefinable :)

(coz a running average approximates a stats parameter: is a model). So plants and bacteria pass, individual protein Cascades pass, thermostats pass etc. you posted since good links! But "agent" is right there with free will and computation: intuitive, useful, and hopelessly ill defined.

I don't want to argue, but as a former neurobiologist I have to point out that while your definition is intuitive, it doesn't define a neat philosophical class. Mathematically, the simplest model of the environment is smth that tracks a running average of a variable and have it affect it's behavior

I have to admit I didn't know your definition of agency, so I had to guess :) With a sentence that long you obv mean "some animals" not "all animals", makng this edge case too easy to defend (gradual emergence). Let it be.

But it's trivial to write a tiny program doing exactly what you described

You lose me already here. I don't know in what way animals have agency in which RNA molecules don't, and in which simple programs (not even LLMs) don't. We diverge at the very basement of your alt-building, I'm afraid. I'm sorry.

I'd say genome (or rather what you get once you let it loose) is the architecture + lots of useful biases. The model is that + the language / culture, so not that different _conceptually_ than LLMs, really.

In what sense is it the same person then? After 100 000 years, what makes them the same human? they speak a different language, have likely changed name, have no biographic memories from 987 333 or 978 444 yrs ago. What then?

If you say "character" then so many pairs of currently living ppl qualify

Another interesting mental experiment are immortal humans (like Icarium in the Malazan chronicles if you read it). Imagine an immortal human. The capacity of the brain is limited. It mathematically guarantees that, like the ship of Theseus, in 100 000 years no traces of original memories survive