iiinova — where your digital self keeps evolving

Back when I started my math undergrad, a professor jokingly told my class, "Double-check your work! When compsci students write silly things the computer gives them errors, but with math, the paper doesn't complain" I realize that, today, the paper *could* complain.

100

Reposted by A. H. Zakai

Cloudflare @cloudflare.social · 28/09/2026

With the new experimental support for the Emscripten target in Rust Workers, many previously unsupported Rust libraries and applications can now be built and deployed directly to Cloudflare’s global Workers platform, including upcoming support for Tokio async. cfl.re/4hlHSNC #BirthdayWeek

blog.cloudflare.com

Supporting native Rust in Workers with the new Emscripten target for wasm-bindgen

0175

A. H. Zakai @kripken.com · 27/09/2026

Thread about claims like "OpenAI should have just sandboxed the agents better" "OpenAI shouldn't have run the agents with the anti-hacking guardrails off" (so many articles are making these points that I'm not going to bother linking)

120

Reposted by A. H. Zakai

noam @noamchompers.bsky.social · 26/09/2026

if you believe the computational theory of mind, the artificial minds debate is like 'what if a *computer* could be.............. a computer'

6928

A. H. Zakai @kripken.com · 23/09/2026

Two points in support of Will's position: 1. Even experts rely on commonplace meanings of words. Say that aliens land on Earth and we communicate with them. No serious person would say, "but we can't tell if they are intelligent without looking inside their heads!" The behavior is enough.

5691

A. H. Zakai @kripken.com · 23/09/2026

Just started wasm-reduce on a 223 MB file ...and that's enough for today

000

A. H. Zakai @kripken.com · 22/09/2026

New release of Binaryen, the WebAssembly optimizer: github.com/WebAssembly/... Among the changes is a new pass I've been working on, ConstraintAnalysis, which propagates logical facts ("equals zero", "less than 5") through the control flow graph

github.com

Release version_133 · WebAssembly/binaryen

120

A. H. Zakai @kripken.com · 22/09/2026

It occurs to me, and this will sound weird, that "AI capabilities will hit a wall / RSI isn't going to happen" people are making a similar mistake as scientific racists are. Both groups think intelligence is like running speed. That it is bounded.

110

Reposted by A. H. Zakai

Vincent Carchidi @vcarchidi.bsky.social · 20/09/2026

[cognitive scientist trying to do physics] now would you say the photon *knows* it's a particle and a wave?

2484

Reposted by A. H. Zakai

Will Stancil @whstancil.bsky.social · 19/09/2026

If you have evidence of a non-mechanical, non-physical component of human cognition, you should go ahead and reveal it so that you can turn all of science since the Enlightenment on its head and transform everything we know about the universe

81826

A. H. Zakai @kripken.com · 19/09/2026

Reposting this comment as the original conversation petered out, but I am curious if I am right or wrong here. Humans have consistently made faster&larger GPUs, and it seems uncontroversial to me that AI models can probably help accelerate that progress? (and if so, this is "RSI")

100

Reposted by A. H. Zakai

Robin Wigglesworth @robinwigglesworth.ft.com · 18/09/2026

‘Greek finance minister encourages Germany to implement reforms’ www.ft.com/content/7a28...

static.klipy.com

Antonio Banderas' Satisfied Reaction

ALT: Antonio Banderas' Satisfied Reaction

1355298

A. H. Zakai @kripken.com · 18/09/2026

My similar take on this is 1. Calculation/computation is one of the few things where simulating it is the same as doing it 2. Calculation is enough to produce all behavior 3. Thinking - calculation <= qualia (i.e., at most: qualia), which is not an insignificant thing, but *very* debatable

010

A. H. Zakai @kripken.com · 17/09/2026

I think a lot of "AI will kill us all" worries came before LLMs proved themselves. I mean, it wasn't clear that AI would actually *be* a mathematical representation of all of human output. Many people had very different ideas of which AI would end up working! It is genuinely surprising LLMs "won".

220

Reposted by A. H. Zakai

Astra ⎔ @astrra.space · 16/09/2026

asking jev "what is the next character in this passage:<...>" with a single choice for each ASCII character

71505

A. H. Zakai @kripken.com · 16/09/2026

Really cool how all it takes to solve Millennium Prize Problems is a Python loop and a chatbot

"The Hugging Face hack isn't a mysterious, supernatural occurrence. It's a Python loop and a chatbot."

3201

Reposted by A. H. Zakai

Pwnallthethings @pwnallthethings.bsky.social · 16/09/2026

Lots of folks say OpenAI's sandbox was negligent but (a) the agents used *several* 0day to break out of the sandbox (b) find me literally any commercial sandbox OAI/Anth have not reported 0day in since (c) Anthropic had an example where they accidentally hacked a security company's malware scanner.

3471

A. H. Zakai @kripken.com · 13/09/2026

Good thread about the possibility of a model copying itself to the outside world. I'd add that I think the reason people are so interested in this question is because of the analogy to biology (virus/animal escaping from the lab), and it's a bad analogy:

130

Reposted by A. H. Zakai

rev. howard arson @theophite.bsky.social · 12/09/2026

the thing is made out of words, so the solution will be made out of words.

41617

A. H. Zakai @kripken.com · 11/09/2026

Sandboxing is indeed a "well-studied security problem since the early days of Java applets", but it very much remains a problem! Literally the best sandboxes still get exploited. I wonder if there was a miscommunication between the expert and the author here.

181

Reposted by A. H. Zakai

Ryan Cavanaugh @searyanc.dev · 10/09/2026

"we haven't invented artificial sweetener because Splenda isn't actually sugar"

1213

A. H. Zakai @kripken.com · 10/09/2026

claude: *proves fermat's last theorem in 13 million lines of lean* me: feels bloated. get rid of all the first-draft cruft. make it fit in the margin of whatever paper size they used in 17th century france

1110

A. H. Zakai @kripken.com · 08/09/2026

There is a funny tension between these two skeptical positions: 1. LLMs are unreliable (stochastic / just predicts words / no meaning / etc.) 2. LLMs only do what you prompt them to (no agency / can't do anything truly new / AI only "hacked" HuggingFace because it was told to / etc.)

030

Reposted by A. H. Zakai

"Online Rent-a-Sage" Bret Devereaux @bretdevereaux.bsky.social · 08/09/2026

Honestly, the biggest surprise to me is that I rather assumed once one fascist ran a country into a ditch - as Putin has done, as Trump is doing - the rest of the developed world would see that and recoil, but it turns out every country is thinking, 'maybe our fascists are just Built Different.'

1639252

Reposted by A. H. Zakai

Rafe Meager (they/them) @economeager.bsky.social · 08/09/2026

this story is crazy. also, looks like academia's not so obsolete after all.

I said that if OpenAI released its result in the way proposed I would go
public with what happened. The reply was, “Why would you ruin your career?”
I replied that I am an academic, and asked why he thought going public would
ruin my career. The reply was, “If you don’t want me to be nice, then I don’t
have to be nice.”

623146

A. H. Zakai @kripken.com · 08/09/2026

To elaborate on this: I think that Wittgenstein is, maybe uniquely, a philosopher that wrote a *lot* about language - that saw language as very, very important - but whose positions would not be shaken at all by the existence of machines that pass the Turing Test. AI wouldn't matter to Wittgenstein

121

A. H. Zakai @kripken.com · 07/09/2026

I could be wrong, but I think Bender's point here is that extraordinary evidence is needed for the scientific claim "LLMs can think". (which is different from "the LLM found a counterexample") That's why she says "we need the inputs". Not because she doubts any specific facts.

230

Reposted by A. H. Zakai

Kieran Healy @kjhealy.co · 07/09/2026

Writing some scientific workflow slides for this week’s MPTC class.

Research workflow for those branches of the sciences where the important thing is to become famous.

1526875

A. H. Zakai @kripken.com · 06/09/2026

+1 Also: "This is fundamentally the same mathematics that is used to predict the weather." ..sure, but Asahi Linux could have made the exact opposite point by replacing "predict the weather" with "prove Fermat's last theorem"

050

A. H. Zakai @kripken.com · 05/09/2026

So OpenAI bots collaborated on multiple external wikis. Do we know how they found each other? In the first case where the message board was inside OpenAI, I can see how they all stumbled upon it: it was in their sandboxes. But do we know how they all found the same obscure German wiki..?

250

A. H. Zakai @kripken.com · 05/09/2026

Confession: I don't understand C++ error messages. Like why is this commit needed on some compilers but not others github.com/WebAssembly/... I am very happy to write+understand the logic of my code but let LLMs handle weird compiler quirks

github.com

ConstraintAnalysis: Parse AND operations by kripken · Pull Request #9075 · WebAssembly/binaryen

Handles this common code pattern: (if (i32.and (A) (B)) (then .. We can apply both A and B in the first arm, and !(A && B) in the second. Also pattern-match eqz(eqz(..)) as that is the ...

110

Reposted by A. H. Zakai

vriska @vriska.bsky.social · 04/09/2026

well it sure is too large to fit in the margin

0392

Reposted by A. H. Zakai

norvid_studies @norvid-studies.bsky.social · 02/09/2026

Send a plagiarism machine... but not for me...

1323

A. H. Zakai @kripken.com · 02/09/2026

The Afghan Whigs have a new album out. 40 years (!) and still putting out amazing music. Like this track: www.youtube.com/watch?v=2EkC...

youtube.com

The Afghan Whigs - The Deepest Part of the Darkest Shadow

YouTube video by The Afghan Whigs

020

Reposted by A. H. Zakai

Vincent Carchidi @vcarchidi.bsky.social · 01/09/2026

The self is a fiction invented by Big Self to sell more inner monologues.

3484

A. H. Zakai @kripken.com · 31/08/2026

+1, reusing the existing words is ok. So long as you understand the differences (and I think most do). Just like it is ok for a biologist to say "evolution finds ways to adapt to harsh climates". Of course, evolution is actually not a person with goals who looks for things and finds them.

020

A. H. Zakai @kripken.com · 31/08/2026

Great thread about how narratives are very useful in interpreting LLM behavior. LLMs are trained to complete texts. That is what they do (and very well). I'd just add that sometimes people (not Colin) think that all LLMs do is "fancy autocomplete" in the sense of "stupid", and that's not accurate:

120

A. H. Zakai @kripken.com · 30/08/2026

This is literally what I imagined in my head in J. V. Jones' Sword of Shadows series. This is a Guidestone

000

Reposted by A. H. Zakai

Grace @gracekind.net · 30/08/2026

If your servers are compromised by a group of agents, don’t say “we are under attack.” Attack implies intent. Instead say, “computer programs are emitting tokens that, when executed by other programs, are causing network requests that are currently pwning my shit”

1642249

A. H. Zakai @kripken.com · 30/08/2026

me, nodding, after recently watching Lost: "the island moves"

000

A. H. Zakai @kripken.com · 26/08/2026

"Could be risky. Yet goal solution."

081

A. H. Zakai @kripken.com · 26/08/2026

Personally I think Bill Gates is right about a tax on tokens. I'd go even further with trainright/learnright laws (parallel to copyright, but restricting training), but if countries aren't going to do that, token taxes are at least something.

040

A. H. Zakai @kripken.com · 26/08/2026

I largely agree with this - humans should understand the code LLMs emit - but it seems like these posts always end up justifying that with "because LLMs aren't as smart as us" var0.xyz/posts/beyond...

var0.xyz

Beyond recall and the illusion of competence — var0.xyz

The important question about AI-assisted programming isn't who writes the code. It's who understands the system.

100

Reposted by A. H. Zakai

Siobhán @shibbi.me · 24/08/2026

“Unlike humans, AI can’t be trusted to give an accurate report of how they work.” Aristotle: “The brain is an organ for cooling the blood.”

611012

A. H. Zakai @kripken.com · 23/08/2026

Searle seems wildly wrong to me here. Has anything since 1984 borne out his position? "Computers do syntax and cannot do semantics" - well, Boden answers him clearly as to how they can. When asked "what is a better approach to study thought than computers?", Searle gives... Freud.

490

Reposted by A. H. Zakai

Hadas Weiss @hadas.bsky.social · 23/08/2026

nephews keeping you on the phone while they play video games is a sign of great respect in their culture

2341

Reposted by A. H. Zakai

Kenneth Black @kennethblack.bsky.social · 21/08/2026

Happened across this unhinged Wittgenstein line

We speak of birds singing. When I was a boy I was bothered in listening to birds, because it obviously isn't singing. Finally someone said to me, "All right, don't call it singing, call it something else"
'. And soon I could listen to birds and enjoy it.

1635281

Reposted by A. H. Zakai

Karsten Konrad 🇪🇺 @zabong69.bsky.social · 19/08/2026

This category most likely includes everyone who ever worked in automated theorem proving, or more general logics. Their club‘s name is Association for Automated Reasoning for a reason.

122

A. H. Zakai @kripken.com · 19/08/2026

I am pleased to report that the John Wick fandom wiki site is called John Wicki

020