Notices by Adrian Cochrane (alcinnz@floss.social), page 11

I am not trying to set up some cardboard villain here, that we can all shoot at.

I am telling you that you and I -- AND I -- need to find the way to change the frames that seem to justify this horrible rig we're all caught up in.

Baby, they don't give a fuck about what intelligence is. They don't give a fuck about the Yale School or neural networks or walks-like-a-duck, or Godel.

They give a fuck -- I am not hating here -- about how important it is that they and their circle stay afloat, in a cold hard world. They give a fuck about extracting value from community because it is the only way they know, it is the way they have found themselves in, it is the way to their identity and value as humans.

When we catch our c-suite commissioning a team to explore CoPilot, what we are discovering is not that CoPilot is intelligent.

We are discovering that the owners have not even the slightest clue of how the non-owners create the value so cheerfully extracted.

When we catch our student copy-pasting GPT answers into their essays, what we are discovering is not that GPT is intelligent.

What we are discovering is that our pedagogy is insufficient to the complex needs of someone less lucky than us as they try to learn. (And sometimes, too, what we're discovering is what a horrible idea certification is.)

When we catch our assistant *insisting* that they be allowed to chat with Eliza in private, what we are discovering is not that a few thousand lines of LISP has created the singularity.

What we are discovering is that we are not taking adequate care of our assistant, and as a result they are sad, and lonely, and afraid to say how they feel.

GPT, regardless of its value as science, is the latest way to extract value from a community's existence and bonds and function and labor and intelligence and skills and talent, and give it to people who are neither in nor of that community.

The thing about engaging in the delicious pilpul of the intelligence conundrum -- oh I feel it, I feel it, I'm an AI dork from way back -- is that it is an excellent distraction from seeing the extractive capitalism that has made and continues to make our lives -- our *privileged* lives -- bereft of meaning.

Oh, oh, let's debate Turing & Searle & Hofstadter & Minsky! Let's be angry, let's be slow, let's be open, let's be, let's be, let's be . . .

let's be not seeing the forest for the trees.

O.K., what's my plan for this week regarding the Argonaut Stack?

I'll go through (basically) all my components & release Text v2-compatible updates. When I come to my language bindings I'll see how much is fixed by constraining bitsizes when dereferencing.

I'll tidy up & fix my CatTrap API.

I'll write a blogpost, & publish CatTrap.

And claim some of my funding...

@PCMag I'm not sure, but it sounds like NVidia's definition of "ethical AI" is fairly shallow. The article focuses on the guardrails put in to reduce the bigotry GPTs spit back at us... Which itself is built on its own exploitation.

On the news of Epic Games aquiring Bandcamp, and possible alternatives: As it depends entirely on my free time (and asides two weeks early this year I've in fact been struggling to make space for it for a while now) I can't make any lofty promises, but I will say that https://codeberg.org/simonrepp/faircamp is still very much alive on my agenda. I received a lot of positive feedback last year (thanks again everyone) and with the Bandcamp takeover there's now more than ever a motivation to bring this towards an official release. Still: It's done when it's done, self-care and a healthy work mode come first. ( ◡‿◡ *)

Web fingerprinting is worse than I thought - Bitestring: https://www.bitestring.com/posts/2023-03-19-web-fingerprinting-is-worse-than-I-thought.html

Regarding prime numbers:
* Return iteration count of a division
* Calculate primes with a given sieve size
* Iterate over primes via some of these other routines
* Probably-prime test with fastpaths
* Factorize a number
* Compute Lucas numbers
* Compute Lucas modulos
* Miller-Rabin primality test

2.5/2.5 for today! I should be able to finish tomorrow on MPZ.

Then random, scan, & tune.

Also there's converters:
* From MPF
* From signed & unsigned longs
* From MPN
* From doubles
* From strings
* To strings (most of the complexity)
* To files

As well as:
* Reallocation (couple)
* Set a particular bit (multiple codepaths)
* Locate first 1 bit (+ & - codepaths)
* Locate first 0 bit
* Select appropriate exponentiation codepath
* Negating a number, possibly with a copy
* Modulo via `mpz_tdiv_r`
* Get, modify, finish, & write "limbs" (digits) field routines

2/3 for today!

Today I'll continue studying MPQ, which largely serves to wrap various MPN routines with memory management & normalization. These include:

* Copying fields from one MPZ to another.
* Random (several wrappers)
* square-root + remainder (couple)
* test perfect-powers
* Modulo (multiple codepaths)
* Multiply wrapped in a loop for product.
* Multiplication (multiple wrappers, including for powers-of-2)
* Powers (multiple wrappers)
* Lowest-common multiple
* Jacobi (& Kronecker) primality test

1/2

@xerz
There has been quite a lot of improvements under the hood in @WebKitGTK, many related with performance and supported web features, but the GNOME release notes don't even mention that the GTK4 support was a major undertaking, and that we have everything lined up for zero copy operation, including accelerated media playback. This teaser was shared some days ago in Reddit: https://www.reddit.com/r/gnome/comments/11pav5t/gnome_web_44_leaps_and_bounds/

Reverse-engineering the Globus INK, a Soviet spaceflight navigation computer - Ken Shirriff: http://www.righto.com/2023/03/reverse-engineering-globus-ink-soviet.html

The Great Gaslighting of the JavaScript Era - Jared White: https://www.spicyweb.dev/the-great-gaslighting-of-the-js-age/

The great JavaScript lie - Go Make Things: https://gomakethings.com/the-great-javascript-lie/

Wow, reading a lot of responses to several new announcements of AI models in my feeds today...

In place of sharing them, let me say: While I am surprised at what these neuralnet models can do, I really can't get excited about them like it feels I'm expected to. Instead I'm much more likely to get excited about neuralnet architectures!

Where the neuralnet's easier to train, as opposed to being pre-trained.

Don't trust AI to talk accurately about itself: Bard wasn't trained on Gmail - Simon Willison: http://simonwillison.net/2023/Mar/22/dont-trust-ai-to-talk-about-itself/#atom-everything

Errors and Zig - Eaton Phil: http://notes.eatonphil.com/errors-and-zig.html