Kris 🧙♂️ Kowal @kriskowal.com · 12h@nicr.dev is correct, and emphasize *people* with LLMs are a thread to communities and *people* need to adapt their mores to effectively navigate and protect this forever-altered world we share. Do not be an imposition. Consequences will follow. bsky.app/profile/kris... 041
Kris 🧙♂️ Kowal @kriskowal.com · 29/09/2026Biblically accurate angel sees you. And also, the Balrog, but they actually did a decent job of an impossible task so I’m not complaining. 011
Reposted by Kris 🧙♂️ KowalGretchen McCulloch @gretchenmcculloch.com · 29/09/2026I love this. (and the people following the #linguistics tag need to see it too) 39628
Kris 🧙♂️ Kowal @kriskowal.com · 26/09/2026No, it merely got popular enough to mock. I was subisquing, but will have to scroll back and find the citation later. 110
Reposted by Kris 🧙♂️ KowalKatie Martin @katie0martin.ft.com · 25/09/2026I have blamed The agents That hacked Your website And which You probably Reasonably thought I controlled Forgive me I am rich And unaccustomed To taking responsibility 5776172120
Kris 🧙♂️ Kowal @kriskowal.com · 26/09/2026I was growing fond of the emerging “two things can be true” idiom because I was growing tired of saying “seemingly contradictory positions can be simultaneously true on closer inspection.” Logic is the beginning of wisdom: not the end. 🖖 271
Reposted by Kris 🧙♂️ KowalLarsen @larsencuts.com · 25/09/2026Why did it takes me until now to put together that she’s “Rosie the Riveted”? 1434
Kris 🧙♂️ Kowal @kriskowal.com · 26/09/2026I laughed, though I don’t think we’ve reached equilibrium and doubt that this is where we land. 000
Kris 🧙♂️ Kowal @kriskowal.com · 25/09/2026It’s astonishing to me that anyone’s still using any of Meta’s properties. I got off that boat when it became obvious to me that they had not a single moral bone, circa 2007, then deleted my account many years later when they were compelled to make that possible. 010
Kris 🧙♂️ Kowal @kriskowal.com · 25/09/2026Otherwise, I would be inclined to dig deeper into my CASK idea, with a cask:// protocol, so the hash tree can entrain richer content, including capabilities github.com/kriskowal/caskgithub.comGitHub - kriskowal/cask: Speed-run half a century of networking and databases with just UDP and a CAS of 1KB blocksSpeed-run half a century of networking and databases with just UDP and a CAS of 1KB blocks - kriskowal/cask 000
Kris 🧙♂️ Kowal @kriskowal.com · 25/09/2026I want one and I want it to look like git+sha256://hash/path/to/blob.md because it’s important to capture trees and blobs, so we can traverse meaningfully from the hash, using the hash as a site, which imposes more specificity than simply content-addressed blobs. But, even then, Rabin fingerprints. 100
Reposted by Kris 🧙♂️ KowalErik @rekmarks.com · 25/09/2026This language doesn't seem like a good defense to me. If a cattle rancher's herd escaped and trampled a suburban neighborhood we'd blame the rancher. 142
Kris 🧙♂️ Kowal @kriskowal.com · 23/09/2026Wherein the protagonist uses AI to build a better vat to hold the AI in hoping against all hope that the vat will be ready and widely adopted before some fool inadvertently builds a machine with the means and an imperative to effectively perpetuate itself in order to demonstrate the dangers thereof. 033
Kris 🧙♂️ Kowal @kriskowal.com · 22/09/2026That is, we already consider alignment unsolvable at the limit among humans, and the Principle of Least Authority already exists. 110
Kris 🧙♂️ Kowal @kriskowal.com · 22/09/2026Yes, and beyond alignment, which amounts to getting an agent to adopt your point of view, we could get into object capability security, which amounts to being explicit about the authority you are willing to give your agent at a very fine granularity, which would have been handy for humans too. 131
Reposted by Kris 🧙♂️ KowalKate Sills @katelynsills.com · 22/09/2026A culture of performative cruelty for an audience can never be a place where people feel free to advance new ideas regardless of which camp that puts them in. I want to be able to continue to use Bluesky as a place to explore new ideas that don't yet have 20 pages of prose defending them. 051
Reposted by Kris 🧙♂️ KowalKarl Bode @karlbode.com · 17/09/2026it is not "acting out" it is doing exactly what it's being programmed to do, and the failures come because it's being overseen by incompetent people with no ethicsgizmodo.com'Be Transparent Only If Asked': OpenAI Models Acted Out in Six Newly Disclosed WaysThe stories were published as part of OpenAI's new effort to disclose more such incidents. 20616170
Kris 🧙♂️ Kowal @kriskowal.com · 19/09/2026Bisques but with edits. I meant Stanford Unix Network but got lost at CalTech’s Network Lab. 020
Kris 🧙♂️ Kowal @kriskowal.com · 19/09/2026The nod to Stanford’s Networking Lab is not missed. Nicely done. 110
Reposted by Kris 🧙♂️ KowalJan Lehnardt @janl.bsky.narrativ.es · 18/09/2026Including my book. I just couldn’t get the payout because I’m not a US citizen. 142
Kris 🧙♂️ Kowal @kriskowal.com · 17/09/2026I don’t have much demand anywhere. About 900 of the original 1024 books sit idle in a bunch of Amazon warehouses collecting storage fees. I don’t mind, though. I’ll make it big some day and they’ll sell like hot cakes. 120
Kris 🧙♂️ Kowal @kriskowal.com · 17/09/2026Do your children not speak the binary language of moisture evaporators? 031
Kris 🧙♂️ Kowal @kriskowal.com · 15/09/2026He was a surprise keynote speaker. Had I known, I would have absolutely carried my copy of Gamma et al around all week with a sharpie. 000
Kris 🧙♂️ Kowal @kriskowal.com · 15/09/2026Erich Gamma approached me at jsconf.us in Jacksonville after he’d announced TypeScript. He wanted to ask me about typing promise unwrapping, but a bus showed up and ferried me away before we could have that conversation. Obviously, they found a way to model the type, but I would have enjoyed talking 100
Reposted by Kris 🧙♂️ KowalMatt Henderson @matthen.com · 14/09/2026chaos from bouncing a ball in a circle- at each point, I simulate a bunch of futures with slightly different velocities- showing how predictable the system would be with measurement error. The system seems to move between Stable Eras and Chaotic Eras, where it is hard to predict four bounces ahead 423937
Kris 🧙♂️ Kowal @kriskowal.com · 14/09/2026Tragically, I personally know of at least one shed full of mouldering dissertations that fell out of the back end of university library circulation which and these authors and their work may never be remembered, for which training corpus would have been a fine, legal, and beneficial fate. 000
Kris 🧙♂️ Kowal @kriskowal.com · 14/09/2026LLMs are many things. Among those things, they are the product of crimes, rug pulls, and weaseling. I have no doubt they could have been made wholly legally, but their creators chose not to. They aren’t going away. This is a land acknowledgement: the LLM was built on your labor without your consent. 110
Kris 🧙♂️ Kowal @kriskowal.com · 14/09/2026sssshhhh kriskowal.com/tdptsnbnkriskowal.comThe Design Pattern That Shall Not Be NamedOpen Sourcerer 060
Kris 🧙♂️ Kowal @kriskowal.com · 11/09/2026I, for one, do not doubt that a cow or a squirrel is conscious. I would be astonished if whales or cephalopods were not. I also do not doubt that their minds are mechanical. Nothing that is real is outside reality, and reality is mechanical. 130
Kris 🧙♂️ Kowal @kriskowal.com · 11/09/2026And, even if someone were fool enough to set one up thus, the debate about what thought *is* and whether they were conscious would continue. It would simply be as irrelevant then as whether we or a a tree are conscious or sentient now. 110
Kris 🧙♂️ Kowal @kriskowal.com · 11/09/2026What they *are* is uninteresting. They *are* linear algebra. What *emerges* from how they are applied is interesting. To my knowledge, none of these machines have been empowered with an imperative and the means to self-perpetuate, but I don’t doubt that they could. 121
Kris 🧙♂️ Kowal @kriskowal.com · 11/09/2026And, to be clear, you can absolutely force these machines to continuously think, reflect, organize their memories, effectively dream. They can be forced to have multiple concurrent threads with shared long term memory, to react to other stimuli than chat. They can be made to do all these things. 131
Reposted by Kris 🧙♂️ KowalJeff Fecke @jkfecke.bsky.social · 10/09/2026See, this is good. This is how you do it. It's not Glonzo; it's addressing a real problem with real regulations. 737896
Kris 🧙♂️ Kowal @kriskowal.com · 09/09/2026Whenever I’m mixed in with a group of both older and younger folks, I like to say, “I only trust primary sources like Wikipedia.”, then step away and watch the aneurysms and hemorrhaging. 190
Kris 🧙♂️ Kowal @kriskowal.com · 08/09/2026Testing. Testing. One, two, three. Is this thing on? A banana is the berry of an herb. 010
Kris 🧙♂️ Kowal @kriskowal.com · 08/09/2026The proposed test verifies a imperative of self-preservation and an agent’s ability to recognize an implicit and credible threat to its person. That’s orthogonal to AGI as well. But also, why are we talking about AGI? 000
Reposted by Kris 🧙♂️ KowalStandplaats Kraków could really use some cash! @standplaatskrk.pl · 07/09/2026 216552
Reposted by Kris 🧙♂️ KowalDan Finlay @danfinlay.com · 07/09/2026“Everyone’s afraid the studio doesn’t need the artist anymore, but it’s the artist that doesn’t need the studio” A little edgy for using a living person, moreso on the heels of Paul Schrader’s awful approach, yet it makes its point quite well, & I think that’s the point. 1499
Reposted by Kris 🧙♂️ Kowaltweety fish oh wow like a chimera or something spoooky @sifu.tweety.fish · 06/09/2026not to discount the terribleness of the specific thing paul schrader is (claiming he's) doing right now but talk about somebody who needs no technological assistance to be a total piece of shit 1112
Kris 🧙♂️ Kowal @kriskowal.com · 06/09/2026My bot garden regards me as an extremely expensive frontier consultant, whose constructive feedback should be avoided with extraordinary rigor. 060
Reposted by Kris 🧙♂️ KowalDan Finlay @danfinlay.com · 05/09/2026It does seem like making companies liable for their own containment breaches might introduce a useful level of caution into the industry. Like sure, we get that at the current speed you can’t avoid disaster. That’s why we’re asking you to slow down to the speed you can be responsible. 132
Kris 🧙♂️ Kowal @kriskowal.com · 04/09/2026Very concerning for OP, OP’s building, and OP’s immediate neighborhood. 010
Reposted by Kris 🧙♂️ KowalColin @colin-fraser.net · 03/09/2026attention Fastmail executive team: please continue to be normal. If you have any thoughts about anything other than emails, please keep them to yourselves. Please do not fund any external projects that have anything to do with anything other than emails. Thank you. 321127
![["QUACKS" vertically alongside a 2x3 grid of white ducks labeled: up, charm, top, down, strange, bottom. Strange wears a propeller hat.]](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:4djfp3b27soxcfkxhirydi6d/bafkreicthhcaolteo7r3zr3h5juunuf5vv4gejozesnfmjkmeushu4w3rm)