iiinova — where your digital self keeps evolving

It also seems odd to me to assert that LLMs do not "understand things" because they don't have the surrounding needs that most biological organisms do. We could construct a mechanical "metabolism" around a LLM that provides it attention and interests, but the LLM itself would remain unchanged.

120

I definitely agree with that, and there's no shortage of ...unsettling anthropomorphizing out there.

000

Not "sufficiently similar" necessarily, but quantifiable in a way that things can be objectively judged against it: bsky.app/profile/craz...

100

Tautologies are inherently true, but they also illustrate nothing. If anything that we assert can be responded to with "that's philosophy at its root" then it means nothing.

100

I agree, but I don't see that as a fuzzy "philosophical" question, it's pretty concrete. 1. Define all the traits that would cause you to think that something "understands something". 2. Are all those things physically quantifiable? 3. Do humans conform to that? 4. Do LLMs conform to that?

000

To be perfectly honest this just feels like "it's still philosophy" in a "turtles all the way down" tautological sense.

100

Well, I deliberately said *if*, so as to not make assertions about the current state of things. But overall I think this is why I just find philosophy so frustrating, it's all so circular and navel-gazy. "These things are perceptibly identical" is a practical argument, not philosophical.

100

But it's commonly accepted that humans are reasoning. If a computer is doing something that is functionally and perceptibly identical to that reasoning, you don't need to retreat all the way into "well how do we know if ANY of us is reasoning" philosophy to ask what the difference is there.

130

I admit it seems very odd to me to assert that something needs a metabolism to "understand something".

160

Chris Carollo @crazytalk.bsky.social · 20/09/2026

How does it ignore biology? Biology for humans, machinery/silicon for LLMs. My read is that Will is asserting that the difference between "understanding things" in those two physical spaces is closing rapidly (or has closed), so people are now searching for non-physical ways to distinguish them.

170

Sentiment among republicans right now is not particularly good!

001

Chris Carollo @crazytalk.bsky.social · 19/09/2026

Fundamentally humans engage with the world via "select next muscle impulse". But obviously that's not where anything interesting or frankly incredible happens, it's just the base mechanism of expression. Similarly, LLMs are "predicting the next word". But that's not where anything amazing happens.

131

Chris Carollo @crazytalk.bsky.social · 19/09/2026

Ok so when it's doing something novel, like a math problem, what is it doing? It sure seems like its (colloquially) understands what it's doing. It's for sure just not doing vanilla next-token prediction on a corpus of data.

100

Whether it "understands" (however you're defining that), or produces behavior indistinguishable from understanding the material, it's not particularly relevant.

100

The worldwide nature of this, with different currencies, cuts against it though.

000

Not at all constrained to the left! "Things are bad now" is pretty implied by MAGA.

120

The big thing is that you'd expect dissatisfaction from all of those things to decline as they recede into the past. But it's done the opposite!

000

Chris Carollo @crazytalk.bsky.social · 17/09/2026

I'm pretty sure you could build a similar list during the '70s too though. We're just not in a uniquely bad time by most of these sorts of metrics.

200

I bet assembly programmers would consider typescript code a "spec" haha

130

Chris Carollo @crazytalk.bsky.social · 16/09/2026

"All modern infrastructure gets permanently disabled" may not be literally an extinction event, but it's pretty close.

100

Chris Carollo @crazytalk.bsky.social · 15/09/2026

Also, weird as it is, 2024 is a *long* time ago in this space.

140

Chris Carollo @crazytalk.bsky.social · 15/09/2026

The fact that intelligence really does seem to mostly just rise from massive scale and speed, is pretty wild. And more than a little humbling.

010

Chris Carollo @crazytalk.bsky.social · 12/09/2026

What evidence do you have that it harms turnout?

030

Chris Carollo @crazytalk.bsky.social · 10/09/2026

My question is, if this wave materializes, is that concrete enough that the narrative is forced to change.

310

Chris Carollo @crazytalk.bsky.social · 10/09/2026

“We will pay you but only if you keep us in power” seems pretty different than standard policy platform of “if we gain the ability to do X we will do X”. The Covid checks were an admittedly weird case of X.

000

Chris Carollo @crazytalk.bsky.social · 09/09/2026

They do have a policy page. It’s…kinda wild.

000

Chris Carollo @crazytalk.bsky.social · 06/09/2026

The best critique IMO is that a lot of people might have switched from “good” to “fair” and the graph would make that pretty small difference look big.

000

Chris Carollo @crazytalk.bsky.social · 31/08/2026

...do you think the people in the 1970s weren't experiencing their now?

1120

Chris Carollo @crazytalk.bsky.social · 27/08/2026

I don't get what's so controversial about this -- no matter WHAT you're doing or what the target is, if you're deliberately winking at racial slurs then it's probably bad. This seems completely self-apparent to me.

170

Chris Carollo @crazytalk.bsky.social · 26/08/2026

I do think that lots of people have used the term in non-winking-at-racism ways. I have used the term jokingly before. *This* particular person is definitely *not* doing that, given their "wireback" usage which is for sure winking at racism. Which gives me pause in using "clanker" again.

000

Chris Carollo @crazytalk.bsky.social · 26/08/2026

"Wireback" is very clearly is a play on a very real slur used against people. Anyone using it is winking at that other use.

110

Chris Carollo @crazytalk.bsky.social · 22/08/2026

That's a pretty wide brush. GiveWell is EA and does lots of good work and is not...whatever this is.

010

"Solving the Elon almost trillionaire problem" is a very different thing than our ability to tax "most rich people" which is what I was replying to, and is in fact the entire point of the OP's quote-tweet.

020

Very few rich people are Musk or Bezos and over-indexing them isn't addressing the real situation. Almost everyone who's rich will get hit by income and capital gains taxes.

100

They also can be unpredictable -- sometimes I'm doing a right-on-green with the crosswalk "don't walk" but it'll instantly switch to walk. Flipping whether I have right of way with no warning is bad! They should work in sync with the lights and be consistent (in this case, always a walk w/my green)

000