&gt;&gt;=

11h

It'd be nice if tech discourse were less informed by the comments of either Ellul or (that damned Nazi) Heidegger.

Setting up a bridge between Matrix and Discord was a huge pain but it works pretty smoothly

therapyspeak tricks people into experiencing their own empathy as a violent infliction and their own conscience as another's manipulation and I think that's pretty psychopathic ngl

I wish there were a twittiquette in the way old forums had some level of normativity cause like the way people even on this relatively tame platform talk to nobody strangers like they proclaim invective indirectly at public figures and companies is unacceptable. You can't talk to people like that.

It's not difficult at all to put the phone down and not chew someone out who doesn't personally deserve it, and the treatment of the day's main character is often downright disgusting. I don't care if they're wrong and stupid bro, there's a right way and a wrong way to tell them that.

I like chess a lot but I'm not very good at it. One kind of frustrating thing is software chess trainers are not great at helping you learn to see mistakes when making them. There's line memorization post-game analysis is flexible, but I want to know what it feels like to catch myself in the moment

This is just me not being any good, I need to drill my openings and learn good defenses and the only way I do that is book in one hand, mouse in the other, play a bunch of lines until I see the matrix. But I would also like to train down my midgame blunder rate and I don't know a good program for it

The obvious solution is just "play more games" but my lack of confidence gets in my way when I'm playing other people and it makes me sad to see number go down. I don't care to be properly rated or whatever I just want to feel like I know what I'm doing when I pull out a board

maiachess.comMaia ChessMaia is a neural network chess model that captures human style. Go beyond perfect engine lines by analyzing games with real-world context, training with curated puzzles, and seeing what players at hig...

I've been using Maia, which is pretty good and surprisingly human, it uses a similar scoring neural network + search algorithm to Stockfish but is trained to model a human player. Really neat software

I've been beating this drum a few times but in discussion of AI controversy, an important dimension of misplaced anthropomorphism is viewing a tensor in a while loop as capable of exhibiting any autonomous influence over society. Every bad outcome of this technology existing is a policy choice.

Individuals and organizations of people remain responsible for the shape society takes, even if the course of events shaping it was a mere reflection of their natural individual and collective self-interest.

We are not totally free to choose the future of society, because we are constrained by the conditions of production we inherit and people with a great deal of prior power tend to conspire to keep and grow it. But neither are we total prisoners of inhuman forces. Every limit exerts finite resistance.

It's true that the social form and the conditions of production (of which every technology is a part) structure historical events and the conditions of life in society. Neither classes, nor technologies, nor the economy, ever actually do anything though. All real events are realized by people.

When "the class" takes action, it is not the class as an abstraction, it is particular social organizations consisting of particular people acting according to their own goals and designs, the outcomes of which we may estimate, but which are subject to everything specific about the people acting.

There is a certain long trend in the political left to speak about social systems and their products that enable certain social acts animately. This can be read as a poetic register, but it's become commonplace to read the register literally. This displaces responsibility in an unreasonable way.

Marxism is often interpreted as "anti-conspiracist" because it explains bad social outcomes without making reference to any specific plan that renders the world a defective and manipulated version of a "good" society. This can go wrong if the action of society is thought automatic.

Technologies of social control are such partly in virtue of what powers they afford to those in control of them, and partly in virtue of what powers those in control choose to exercise. Twitter was bad before Musk bought it, but thumbing the algorithm to be a 24/7 propaganda reel was his choice.

The agency to use those powers to specific ends, or to use any of them at all, rests with those who dictate how the powers are used, not their mere existence. Affordances shape the possible, they do not dictate action, although much can be predicted from the knowledge of affordances and incentives.

Not to bash Ellul for those inclined to animistic mysticism about modernity as such but matrix multiplication spitting out words into a chat window is not fascist, it's being made by people, many of whom are fascists, with fascist intent. Very different thing.

I've seen hell and it has recursive homotopy

Minor Mobius@minormobius.bsky.social

Yeah the video upload problem has fixed itself so the non link clickers can get disoriented too

ies*, I guess

I wasted 45 minutes trying to find a shorter way home than a 20 minute wait for a bus, finding out my connecting train stops running at 11:30, and having to catch the first train heading the opposite way back to wait for the bus I was trying to avoid, AMA

I used one of those silly analysis tools and most of the most similar accounts to me are men. This thing thinks I write like I man. I'm an absolute shambles.

This is like that time I did one of those and it tried to guess my political views and the damnable thing called me a Republican. I will never recover.

Note on the use of the term "agent": The traditional meaning of "agent" in the AI field is just a system that takes in real time input and is equipped to act in an input-dependent manner, without requiring any direct input during execution. The connotation is not anthropomorphic, typically.

>>=@noologist.bsky.social

From a regulatory point of view, the "rogue agent" problem is just an expected failure mode of RL that Amodei himself wrote some fundamental papers about. If you build a system that's optimized to achieve goals, there's little to be surprised about when it fails to observe all reasonable constraints