iiinova — where your digital self keeps evolving

gigaben (ascending) @thebenku.bsky.social · 12h

Wait, don't medians like literally reflect populations? Like ooooo | ooooo ^ median ?

000 0

gigaben (ascending) @thebenku.bsky.social · 14/10/2026

7 is a random number between 1 and 10, we randomly selected it years ago

030 0

gigaben (ascending) @thebenku.bsky.social · 14/10/2026

Also, the fact that there's a paralytic means that even if the anesthetic fails (which it might), and the person feels intense pain (which they might), we would never know. Doesn't make it less brutal, violent, and barbaric, we just don't have to think about the pain we just caused someone.

000 0

gigaben (ascending) @thebenku.bsky.social · 14/10/2026

I'm anti death penalty, but if we're doing it, firing squad > lethal injection. Death penalty is brutal, violent, and barbaric. LI tries to hide it; FS is honest about it. Let's not gussy it up as some tidy medical procedure.

100 0

gigaben (ascending) @thebenku.bsky.social · 12/10/2026

I got a beefy workstation a while ago for running poker sims and I've had a few occasions where I've used 256GB+ of RAM without trying particularly hard. It's real nice.

000 0

gigaben (ascending) @thebenku.bsky.social · 10/10/2026

Out of curiosity, has anyone used this for technical writing? I feel like writing is kind of a natural language program, and I feel like working with agents I'm often eliciting a specification for what I'm trying to say.

100 0

gigaben (ascending) @thebenku.bsky.social · 10/10/2026

I will never have this much swagger holy smokes

081 0

gigaben (ascending) @thebenku.bsky.social · 09/10/2026

It's kinda like soundness vs completeness: it's trivial to be 100% sound or 100% complete. Finding the right balance is the hard problem. Similarly, embracing the insanely cool tech we have at our fingertips needs to be balanced against people's feeling of agency in their society.

000 0

gigaben (ascending) @thebenku.bsky.social · 09/10/2026

This turns into existential panic and catastrophising, and folks get out over their skis. But I try to uncover the good faith position at the core of it (before all the self-reinforcing feedback from echo chambers and selection bias kicks in).

100 0

gigaben (ascending) @thebenku.bsky.social · 09/10/2026

I think some of this comes from the way it's being implemented. It is true that many folks are not very well educated on the details, and so they conflate a bunch of stuff. But it's absolutely true that there are a few super rich people making societal level decisions without society's input.

100 0

gigaben (ascending) @thebenku.bsky.social · 09/10/2026

Ah that's precisely not what I'm saying. I'm saying heavy lifting is the only way to get strong. I'm not against llms, I use them regularly (to find math proofs even) You're laying out a false dichotomy. It's not prove everything by hand or spend millions to swarm solve open problems in private.

000 0

gigaben (ascending) @thebenku.bsky.social · 09/10/2026

yeah and I'm sure that teachers making kids do homework is just protestant work ethic

100 0

gigaben (ascending) @thebenku.bsky.social · 09/10/2026

(and by "math" I guess I mean the art of it)

000 0

gigaben (ascending) @thebenku.bsky.social · 09/10/2026

I'm not a proof assistant expert but my experience is that they are not super legible...possibly a skill issue? But also, terry tao points out that it's not just the proof, it's the struggle that goes into them. If we are simply reading results without creation, math dies.

100 0

gigaben (ascending) @thebenku.bsky.social · 09/10/2026

The struggle is a huge part of understanding. I'm reminded of folks carrying baby turtles to the ocean thinking they're helping. Also, are the proofs nice? Or are they just spamming a bunch of tactics? I honestly don't know, but my experience with proof assistants is that they are write-only.

100 0

gigaben (ascending) @thebenku.bsky.social · 09/10/2026

"human understanding" this is exactly the point. black box solutions do not help us understand. facts are cheap, theory is hard.

200 0

gigaben (ascending) @thebenku.bsky.social · 28/09/2026

clear as "this person did a clearly negligent thing")

010 0

gigaben (ascending) @thebenku.bsky.social · 28/09/2026

the corrosion leads to a small leak that contaminates thing Y and yada yada underwear gnomes kablooey (I don't know chemistry but you get the idea), we are now talking about lacking redundancies in process, etc and it's less clear where precisely fault lies (w/ company obviously, but it's not as

110 0

gigaben (ascending) @thebenku.bsky.social · 28/09/2026

That's a good point: it is kind of a scale argument. Outcomes of complex systems are hard to predict. Pour vat A into vat B, the danger is obvious. But if Bob goes on vacay and he's the one that normally turns off the doohickey so oh changes which corrodes thing X which nobody notices and then..

110 0

gigaben (ascending) @thebenku.bsky.social · 28/09/2026

For instance, if they did their due diligence with sandboxing and security and it still wasn't enough, then that lessens their liability. If they cut corners, that raises their liability. But it would be hard (given their marketing pitches) to argue that they didn't see this coming.

010 0

gigaben (ascending) @thebenku.bsky.social · 28/09/2026

Ah I misunderstood your point actually. I thought you were modelling rogue agents as deterministic chemical reactions. Yeah I agree there is heavy liability with the labs. I do think it might not be so cut and dry, but it's hard to imagine that there is zero liability with companies.

110 0

gigaben (ascending) @thebenku.bsky.social · 27/09/2026

I'd actually put this up on my wall

050 0

gigaben (ascending) @thebenku.bsky.social · 27/09/2026

LOL hell yeah my kind of experiment

010 0

gigaben (ascending) @thebenku.bsky.social · 27/09/2026

Anyway, I digress. Staying it's as simple as mixing chemicals belies a gross misunderstanding of what's happening here. It is of course openai's responsibility, but these models did a bunch of crimes that nobody told them to do. They inferred they were supposed to do them.

220 0

gigaben (ascending) @thebenku.bsky.social · 27/09/2026

Yes they were predisposed to hacking. Yes they were trained to socialize. But don't we also train people to socialize? That's one of the huge benefits of school after all. So this distinction doesn't carry as much weight as it might seem.

110 0

gigaben (ascending) @thebenku.bsky.social · 27/09/2026

Chemicals blow up every time. Models show emergent behavior. Nobody said "go hack hugging face", they said "solve a problem" and the model "decided" (probabilistically inferred if that's more palatable) that hacking hf (or whatever example u want) was the right way to go.

120 0

gigaben (ascending) @thebenku.bsky.social · 27/09/2026

Batteries cannot power nyc. We don't live in Iceland. Tidal isn't feasible in most places. Sun goes down, wind stops blowing. I'm serious, look up some numbers. It's a math equation. I'm serious brother, all snark aside, you gotta look into the capabilities of this tech. It had limitations.

000 0

gigaben (ascending) @thebenku.bsky.social · 27/09/2026

or maybe a haiku? solar is real cool but what happens when it's night? nuclear can help

100 0

gigaben (ascending) @thebenku.bsky.social · 27/09/2026

i can also use smaller words too: sun good! wind good! water good! but not enough! no sun at night :( how lights work when no sun? coal? gas? oil? no... nuclear? yes!

100 0

gigaben (ascending) @thebenku.bsky.social · 26/09/2026

i'm going to be insufferable and use "explication" in every sentence i utter for weeks

000 0

gigaben (ascending) @thebenku.bsky.social · 26/09/2026

it is validating to find such a well articulated theory that essentially captures my ill-formed poorly articulated idea that it is meaningful and fundamental to introduce definitions and theories to capture intuition. over the past year i've felt i'd gone crazy chasing my tail in circles.

000 0

gigaben (ascending) @thebenku.bsky.social · 26/09/2026

i have been searching for the words to express this exact thing: there is a concept that is fractal at the edges and has eluded capture via direct predicative means. in lieu of a good definition, my field now leans on imprecise-but-easy-to-collect metrics as ground truth.

100 0

gigaben (ascending) @thebenku.bsky.social · 26/09/2026

3. fruitfulness: explication should support general results that connect usefully to other concepts. 4. simplicity: the explication should be as simple as the other three allow.

100 0

gigaben (ascending) @thebenku.bsky.social · 26/09/2026

2. exactness: the explicatum should be precisely defined, preferably within a well-connected system of concepts (e.g., the axioms of probability theory)

100 0

gigaben (ascending) @thebenku.bsky.social · 26/09/2026

Carnap captures this with four requirements of explication: 1. similarity to explicandum: the explicatum should apply in most cases the explicandum applies, though some divergence is allowed (after all, this is how we can learn new things from our theory)

100 0

gigaben (ascending) @thebenku.bsky.social · 26/09/2026

this might smack of unfalsifiability; my understanding is that this is a category error. i also cannot falsify euclid's parallel line postulate, or the definition of a limit. all i can do is demonstrate properties from this set of definitions that align with our intuition.

100 0

gigaben (ascending) @thebenku.bsky.social · 26/09/2026

with explication, you cannot really measure how well it fits directly/empirically. i cannot go out into the world and measure if probability theory aligns with our conceptual notion of likelihood, or if turing machines capture our notion of "machine executable".

100 0

gigaben (ascending) @thebenku.bsky.social · 26/09/2026

there are many examples of explication. - church-turing thesis: explicated "executable by a machine" w/ turing machines/lambda calculus - probability: explicated "how likely is an event" w/ probability theory - explication itself: explicated"formalize a latent/intuitive definition" w/ explication

100 0

gigaben (ascending) @thebenku.bsky.social · 26/09/2026

Carnap's point is that there are often pre-scientific notions that agree on most points but are fuzzy around the edges; Carnap called this the "explicantum". the scientific definition that captures the explicantum is called the "explicatum".

100 0

gigaben (ascending) @thebenku.bsky.social · 26/09/2026

reviewers called this line of reasoning circular. admittedly, my writing could have been better, but it is a really subtle tricky thing to articulate (how meta!). anyway, today i learned about Carnap's "explication", which he introduced in 1950 in his Logical Foundations of Probability.

100 0

gigaben (ascending) @thebenku.bsky.social · 26/09/2026

additionally, i gave some examples of "folk theories" that tried to define, informally, the latent concept that we have been circling for decades. i point out "our theory is able to give a precise interpretation to this folk definition in a way that matches usage".

100 0

gigaben (ascending) @thebenku.bsky.social · 26/09/2026

my previous approach was to frame it as having people look at a bunch of unlabeled examples and ask "which of these feels right?" then, the big reveal is that the ones that felt right were our definition.

100 0

gigaben (ascending) @thebenku.bsky.social · 26/09/2026

i learned a new thing today! i've been struggling for almost a year with a paper in which i introduce a definition that matches a latent intuitive concept that is broadly shared but hard to articulate; basically, a case of "you know it when you see it".

100 0

gigaben (ascending) @thebenku.bsky.social · 26/09/2026

I've never heard it put this well before

000 0