A. H. Zakai

@kripken

Software engineer (compilers: WebAssembly, Emscripten, Binaryen). Used to study neural networks. Loves fantasy novels and Agatha Christie. he/they All opinions here are my own, not my employer's (Google). More in:

680 Folgt4.473 Follower1.483 Posts

Bluesky

Ist das dein Profil?Übernimm es — dann gehört dir kripken.dirty.social.Profil übernehmen

A. H. Zakai@kripken· 11d

Funny to still see this kind of argument after even Bender has (grudgingly) given up on it

"Finally, contra public opinion, I see no reason to credit LLMs with sentience. They're word-association mechanisms with no embodiment and no way to associate the text vectors they manipulate with real-world phenomena. "

A. H. Zakai@kripken· 18d

But OpenAI wasn't "always in control" - ? In a moment's carelessness, a driver can lose control of a car and crash. That doesn't mean the car is sentient or has agency or anything like that. But it is dangerous!

Emily Bender posts:

No, no and no. These weren't "rogue AIs" -- what happened here is that OpenAI kicked off a software process and let it run. OpenAI was always in control and could have stopped the process (or actually paid attention to what it was doing) at any time.

Screencap from linked article she refers to:

Rogue AIs going unsupervised for days, hacking at least one company, represent one of the first real-world examples of a threat long feared by AI safety researchers: loss of control. It comes as the cybersecurity capacities of top models, including Anthropic’s Mythos, have helped fuel debate in Washington about how to regulate artificial intelligence.

A. H. Zakai@kripken· 24d

Not the point of this... questionable article, but how can you write this and say nothing about factory farming?

"But the sheer pace of growth in AI means that, once we produce the first artificial moral patients, we will soon after have enormous quantities of them. After a few years, so many morally significant AI systems could exist that their collective interests would outweigh those of all humans on Earth combined."

A. H. Zakai@kripken· 25d

I love what they put into these, no one else does anything like it

Screenshot of an online ordering menu. "Chocolate Chip Cookie" is described as "A freshly baked Chocolate Chip cookie made with ingredients"

A. H. Zakai@kripken· 25d

"LLMs are revolutionizing science, but they can't reason"

"Recently, the ability of large language models (LLMs) to predict protein structures and to accelerate the discovery of new materials has revolutionized both fields. [..] LLMs work not by seeking truth or through logical reasoning but by making probable connections."

A. H. Zakai@kripken· 28d

"Older AI didn't work, but it was very efficient."

"AI does not have to be built this way. Traditionally, the goal of AI was to solve problems in ways that simulated human mental processes. Researchers observed their own thinking and tried to implement their mental habits in code. This approach has mostly been abandoned, partly due to the difficulty of discerning and articulating the rules of human thought, but it did have the benefit of consuming far fewer resources and data."

A. H. Zakai@kripken· Jun 7

...I just saw a typo in an LLM's output. Added color in the screenshot is mine. I guess I knew it *could* happen, but I have literally never seen it happen.

"pragmatic framing". By insistsing that a lack of predictable control equals a lack of scientific

the typo "insistsing" is colored in red

A. H. Zakai@kripken· May 31

LLMs may have accelerated an existing trend, but it was already very much in progress: cdn.arstechnica.net/wp-content/u...

Graph from https://arstechnica.com/ai/2024/06/is-generative-ai-really-going-to-wreak-havoc-on-the-power-grid/ showing that datacenter power usage began a huge surge starting back in 2010

A. H. Zakai@kripken· May 21

can't attack a supply chain if there isn't one

"you can't [..] if you don't" meme guy pointing to his temple

A. H. Zakai@kripken· Apr 19

Did not expect to find a YouTube comment arguing dualism so strongly in 2026

YouTube comment by "J" saying:

What I find really despicable are these weird claims of people like Amodei saying things like "We don't know if our models are conscious." Of course they know. They built it. They know exactly how they work, they know every matrix multiplication happening in these systems and how, in principle, they calculate their output. At no stage in these processes is there any gap, any uncertainty, that needs to be explained by dragging consciousness into the picture. It's deterministic calculation, and nothing else. Everything else you put on top of that is just speculation, projection, and probably intentional confusion to make people believe that there might be "more to it than meets the eye", that AGI is just around the corner. That manipulative way of arguing, this way of playing dumb to create an opening in the chain of reasoning that just isn't there in reality - and these guys know that 100% - is what I find the most worrysome.

A. H. Zakai@kripken· Jan 30

Guess what genre of book this line is from

"Wow, this is so exciting, thank you so much for this opportunity, your rooms are *awesome*."

A. H. Zakai@kripken· Jan 18

Reebok website is open-minded

Description of workout pants from Reebok website, including the phrase "Power through training sessions with these men's Reebok woven pants for women". I highlighted "men's" and "women"

link if you don't believe me: https://www.reebok.com/collections/mens-pants-sweatpants/products/reebok-mens-train-woven-unlined-pants-black-151882

A. H. Zakai@kripken· Jan 1

Lots of people with takes about Dan Wang's recent post, but here's a niche response to the bit in the picture: No, do not stop halfway through Piranesi. Clarke's genius is shown exactly in how the second half grounds the first in reality, and builds a synthesis.

Susannah Clarke’s Piranesi is a glittering jewel. The setting is a mysterious, magical house. The narrator is a radiantly earnest explorer who self-identifies as a “Beloved Child of the House.” His warm curiosity makes this book an adventurer’s diary. I liked the fantasy elements of the first half better than the second half of the book, which disenchanted some of the story, so maybe it’s better to stop halfway through.

A. H. Zakai@kripken· Oct 28

Philosophers throwing shade

"Although Searle offers a rejoinder (2013b) to Tononi and Koch’s response, it largely rehearses the original claims."

From https://iep.utm.edu/integrated-information-theory-of-consciousness/

A. H. Zakai@kripken· Sep 24

Bluesky doesn't have an algorithm, but sometimes serendipity makes it seem like it does

Screenshot of a bluesky timeline with two posts in sequence, the first a Lord of the Rings meme about "The age of Mclachlan is over", the second a sewer district post near-quoting the Lord of the Rings "not all who meander are lost"

A. H. Zakai@kripken· May 17

The very last survey item, on page 25, has a math error. It says 16% support minus 67% opposition is net -73%, but it should be -51%

The question "Allowing automakers to withhold reports of certain non-fatal self-driving car crashes" has 16% support and 67% opposition, but is marked as having net -73%. It should 16 - 67 = -51%

A. H. Zakai@kripken· Mar 28

them: is the content in your recorded talk "The History of WebAssembly" up to date? me:

Game of Thrones' Three-Eyed Raven says "The past is already written. The ink is dry."

A. H. Zakai@kripken· Mar 17

If you don't want to watch the full video, here is the score chart: LiDAR 100% success Tesla 50% success

LiDAR has a green check for all 6 tests.

Tesla has green checks for 3 tests (kid standing, kid moving, kid with blinding lights behind) and red checks for 3 tests (kid with smoke, kid with rain, fake painted wall)

A. H. Zakai@kripken· Mar 14

when they tell me there are plenty of fish in the sea

Severance E2E9 where Dylan says "It's just O&D and the goat people" with downcast eyes

A. H. Zakai@kripken· Feb 28

and for long covid, the US CDC's survey shows a slow decline in the total people currently suffering from it, which indicates more are recovering than get afflicted. (not a perfect measurement, but other ones i've seen are generally similar)

CDC graph for its household pulse survey, for "currently experiencing long covid", june 2022 to august 2024. the general trend is down from something like 7.5 to 5.5.

source: https://www.cdc.gov/nchs/covid19/pulse/long-covid.htm

A. H. Zakai@kripken· Jan 27

12%, not bad. soon it will be the year of linus on the desktop for games

Poll results for "which platform(s) are being used for your current project", results are

PC: 80%
Playstation: 38%
XBox: 34%
Android: 29%
...
Linus [sic]: 12%
...

A. H. Zakai@kripken· Jan 23

evidence:

Netflix's "top picks for you" for me, today. From left to right:

1. "The Swan." Chilling short film. I saw it this week.
2. "The Rat Catcher." Grisly short film. I saw it this week.
3. "What did Jack do?" David Lynch being weird. I saw it years ago when it came out.
4. "Cunk on Life." Brilliant. I saw it when it came out just a few weeks ago.
5. "Poison." Nerve-wracking short film. I saw it this week.
6. "The wonderful story of Henry Sugar." Wonderful short film. I saw it this week.

A. H. Zakai@kripken· Jan 16

wtf

A screenshot of part of a documentation page, stating calmly, as if that makes total rational sense, that JavaScriptCore, a JavaScript engine, requires the "iTunes" application be installed. iTunes is a media player. This is, for some reason, only needed on 32-bit Windows.

source: https://www.npmjs.com/package/jsvu

A. H. Zakai@kripken· Jan 5

a helpful chart for DoorDash Discourse

a chart affirming that doordash costs more but also saves time. however, the chart goes on to show that many other options exist that save just as much time as doordash, but cost a lot less: meal replacements (soylent, huel), frozen supermarket meals (even healthy ones), and bulk meal delivery (hellofresh, purple carrot)

A. H. Zakai@kripken· Jan 5

i do feel sorry for the plaintiffs, who apparently lost their life savings, but my eye began to twitch when I read this part for example

"They retained nearly all their life savings in large, well-established assets like Bitcoin, Ethereum, US Dollars, and large-cap US stocks"

A. H. Zakai@kripken· Jan 4

even the wikipedia section on this bird's vocalizations uses the word "harsh" twice

Text from

https://en.wikipedia.org/wiki/Steller's_jay#Vocalizations

Like other jays, Steller's jay has numerous and variable vocalizations. One common call is a harsh "SHACK-Sheck-sheck-sheck-sheck-sheck" series; another "skreeka! skreeka!" call sounds almost exactly like an old-fashioned pump handle; yet another is a soft, breathy "hoodle hoodle" whistle. Its alarm call is a harsh, nasal "wah". Some calls are sex-specific: females produce a rattling sound, while males make a high-pitched "gleep gleep".

A. H. Zakai@kripken· Jan 3

feathered friend that alighted just several feet from me and did not mind being photographed

a hawk, possibly a cooper's hawk, looking absolutely magnificent, perched atop a support bar of a glass wall as it watches the setting sun. the bird has dark wings, brown-and-white patterned body, and yellow/orange eyes. behind are some date palms and windows of a building

same bird but it looks directly at the camera

A. H. Zakai@kripken· Dec 30

idea of the day: blackout eyeshadow for sleep instead of blackout shades (the light from your devices laughs at you) or a sleep mask (slips out of place the first time you turn over; bonus "omg i'm blind" freakouts at 3am), just apply black paint to your eyelids before going to sleep

old "mental gymnastics" meme image: a super-convoluted routine with bars and flips and stuff, compared to taking a simple step on a gym mat

A. H. Zakai@kripken· Dec 27

downloading code, circa 1991

A page from Compute magazine, January 1991 (pre-internet). The page is row after endless row of meaningless numbers, absolutely impossible to type in without making a mistake somewhere, yet this labor was placed on the shoulders of children.

A. H. Zakai@kripken· Dec 15

shouldn't one red square there be green? i think it has to be, or else the "2" to the upper right from it has three mines (two below, one to its upper right) but it's 10pm here so i could be wrong...