aevaI've been thinking a lot lately about the art education I got 20 years ago, and the importance that was placed upon learning how to really see what is in front of you, how to deconstruct the technical aspects of other works, looking at a lot of art, collecting good reference images, and talking about art.
This process is really important for developing and maintaining the technical skills of working in traditional media, because if you are not able to establish cultural and observational reference points you lose the ability to "see" as you work. Like, after you have been working on something for several hours everything sorta looks correct and wrong at the same time, and if you don't have something to ground it you'll drift in ways you don't want to.
It is also a really important set of skills for learning, because you can go to an art museum and look at works that inspire you and figure out (or make an educated guess at) the technical processes that their author used to make them, and then you can apply your learnings to making something new. You don't go to the museum to find things to copy without understanding.
One of my peers from that art program once remarked to me that she's seen interesting things happen in the works of artists that either don't have this training or only reference art made by other amateurs - work that is not anchored in either a cultural tradition or naturalistic observation or both tends to drift in strange ways
the example she gave was a self taught artist who only referenced manga, but was not familiar with the techniques or anchors the mangaka they were copying themselves knew, and so said artist's style ended up exaggerating all of the stylistic elements of the manga they were referencing
my point to all of this is, I don't at all believe you can look at / watch / read / listen to something and not be changed by it even in a small way. the human brain is a hungry for patterns to learn from, and everything you learn is eventually pruned or mutated through continuous re-encoding. you need to keep absorbing high quality examples and information throughout your career to preserve your skills. I think this applies generally.
this is why I think slop is so damaging, especially so as it gets harder to spot, because you're feeding yourself with vacuous garbage that superficially resembles information. if you accept it as valid, every related skill that you've worked hard to sharpen up to that point is fundamentally at risk because as they get re-encoded they will be adapted to accommodate the miscategorized noise.
now, not everything you see in the world is going to help your career as an artist, and I think it's true that there's plenty of things you can internalize that are destructive to your ability to make the things you want to make. that sounds pretentious, but I don't think it is: a simple example is internalizing the idea that you shouldn't make art because you don't already have the skills to do it with perfection is effective at preventing many from ever developing those skills.
anyways, my conclusion from this right now is it may be a good idea as a professional programmer who wishes to retain her hard earned skills, to make a regular habit of going to the museum as it were~~regularly reading the source code of successful projects that have non-superficial high standards, understanding how it works, why it works, why it is the way it is, and also looking at the history of how it grew over time.
I say "non-superficial" because strict adherence to a random grab bag of engineering best practices is not holding yourself to high standards if you don't understand any of them. That's just posturing.
Fabian Giesen@aeva These last few years have broken something deep within me and I have no idea what to do with myself now.
It's not about the generative whatever particularly, that's certainly not helped but it's mostly yet another manifestation of something that's been eating at me for way longer.
@aeva I want to understand, and to be understood.
That's not the only thing about me, but it's a core tenet about my personality and my construction of identity.
Almost everything I do flows from that in one way or another.
That's why I default to these long, rambling info-dumps, for example.
@aeva Probably my most formative single childhood experience is, over and over and over again, running into and beyond the limits of what my parents were able to explain to me (they really tried, but especially pre-internet, which this was, you run into a wall of how many difficult research-requiring questions of a precocious child you can actually answer in any given day), and likewise, my continuing inability to explain what's in my head.
@aeva My parents were gracious about this, other adults in my life, less so.
Shout-out to the teachers at my nursery school who decided that a 4yo couldn't possibly actually want to know how a differential transmission worked and that therefore I must be some attention-whoring little bitch who loves to pick up words I don't understand and who should really shut the fuck up. (Not in those words, mind - this was a CATHOLIC pre-school! - but the sentiment was there.)
@aeva er, lots more to unpack there I guess (not for here!) but suffice it to say that the preemptive infodumping is a tic I picked up right around that time (I was verbal pretty early, but the infodumping started then - there's "video evidence" from my uncle, who had a camcorder. Very clear diff between me before/after entering nursery school).
So, er, yeah. Some damage there. Anyway, the over-explaining/desperately wanting to be understood has _deep_ roots.
@aeva More to the actual point, like it or not, this has just turned into a schema of what I look for in personal relationships.
(It's just been really hard all my life to find anyone who can actually follow along when I get going.)
Anyway. What does this have to do with the current mess? I'm getting there, but first another, much more recent episode.
@aeva This was a few years ago, on Twitter.
A mutual, who I'd met in person, replied to me posting a link to a new blog post with something to the effect of "Oh great! Always love your posts. Usually don't understand a word though."
They thought they were giving me a compliment, what they actually gave me were the seeds of an existential crisis.
And I know it was hyperbole, but the "don't understand a word" has been living rent-free in my head ever since.
@aeva It's one thing if at some point two thirds through a post I go off on some tangent that matters to me but that 95% of readers don't care about.
But "don't understand a word"? I really, _really_ try to make my technical writing as clear as I know how to and if the end result is, apparently, incomprehensible gobbledygook, then what the fuck is the point of writing any of it in the first place?
@aeva So that's _that_ light-hearted tangent.
Now, finally, on to my actual point.
As alluded to in both of these digressions, I deeply care both about understanding what I am doing, and about being understood.
Often to my detriment. I shouldn't care as much as I do. It's not something I can turn off. And like it or not, it's directly intertwined with my need for human connection.
@aeva Figuring out a solution to a problem, having a seed of understanding within it, and getting to share that seed with others is, not to mince words, the only reason I put up with any of this shit.
A friend mentioned his pet theory a while back that shepherding LLMs sucks for people who like programming because that means now they have to be managers and that's a different skill set.
@aeva That's not _entirely_ wrong but it's missing the point by a mile if you ask me.
It _is_ a different kind of activity, but it's not "management" either.
You get to play-act as the world's worst micro-manager constantly telling your "agents", who never learn a damn thing, to try again.
Mentoring or managing somebody (hopefully!) involves some kind of development where they learn something and grow into their position.
@aeva That's not what's happening here.
You're in a groundhog day loop with an intern with infinite stamina, ADHD and hopped up on 3 cans of Monster Energy cranking out piles of code between bouts of tachycardia, and every hour they hit the neuralyzer and need everything explained all over again.
I don't know what ring of hell exactly this is, but it _is_ a ring of hell, and it's not "management" by any sane definition of the term.
@aeva Mind, I'm not actually doing any of this stuff. But. BUT.
There are _so many_ people around me who are, apparently, dead set on treating the scenario I just described like it's a desirable outcome instead of a dystopian nightmare, and actively working towards it.
I guess if you treat programs as some necessary evil that's a speed barrier between you and all your glorious plans working out, there's some sense to that.
@aeva It just so happens to be that if you're some misguided rube like me, apparently, who cares about how programs can be some concrete embodiment of entirely abstract ideas that you could otherwise never communicate to anyone, this is just casually shitting on your life's work and going "yeah who needs THAT".
And I _hate_ how this sounds melodramatic and petty but I have no other words for it.
@aeva This is not about the tools themselves. There's _so many_ problems there, and much has been written about it, but that's not what I'm getting at.
The thing that's _really_ getting to me is just how much of the SW world, including so many people around me, is going "oh yes, finally" about this.
I've always felt pretty alone in my caring about things the way I do but was telling myself that I did find my people who care about things the same way in the end. Evidently, no.
@rygorous i feel every word of this so hard, and i just want to say i cherish your infodumps and i know exactly the alienation you are feeling from all of this. i've learned so much from you both directly in our conversations and from reading your writings, and i hope you never stop.
@rygorous i'm incredibly disheatened by how many career programmers are ready to just give up and live a simple life feeding coins into slopmachines. i'm... not optimistic that people will change until it all breaks, and i would be astonished if any of them took responsibility for it when it does, buut i am also heartened by the good many people here who want to burn down the darkness factory
Andy Robbins
skategoat 🐐 🇵🇸@ashalah @rygorous there's a very real possibility that the software engineer salary will deflate significantly possibly all the way down to a minimum wage if the chat bots improve such that most programming careers are reduced to a few chat bot supervisors (or if the money decides it doesn't care that they're not fit for purpose and does it anyway). you wont be able to support your family on that, and you should come up with a plan for that *now* and not after it happens.
@ashalah @rygorous that's the singular obvious outcome that will happen to all of the "join us or be left behind people" and everyone they dupe if they end up being right. the chat bot chaperone is not going to have a 2010s engineer salary, and the chat bot chaperone industry is going to have far fewer seats than the ever shrinking tech industry has today
@ashalah @rygorous now, it's possible that you'll eek out a living as a specialist in an esoteric corner of the field as a graphics programmer. of course nvidia's finest are working on that particular wrinkle around the clock so who knows. personally I hope they fail horribly and embarrass themselves
TommyTorty10At the risk of making a major fool of myself, I don't think many software engineer salaries will get nearly that low. If the past few years of chaos and layoffs couldn't do it, I think the sort of thing that would, would have farther reaching consequences that would shake the whole world enough to cause a big economic crisis. Could that temporarily destroy salaries? Yes, I think so, but I don't think even the billionaires could live with a crisis that large for very long. I also think that really skilled labor will still cost quite a bit.
I think about my university music instructors, and how they make more money playing downtown on weekends than teaching. Sure, it's really hard and uncommon to become a successful musician, but those who are, still make good money. I think that's sorta what we might see. Software engineering might become even more competitive and selecting for people who can wrangle executive delusions and also produce quality products without toasting a company.
I think we're well past the point of diminishing returns for llm development, at least from a monetary perspective. Companies sank Billions and Billions of dollars and scraped the web before it got polluted with AI, and imo, the last several generations of LLM's have only gotten a little bit better at a time. The return in quality just isn't matching the massive increase in investment, so I don't think there's gonna be a major jump in quality around the corner. I think we could see other types of models get greatly improved if they get similar investment. For example, if somebody pumps a billion into a model hooked up to a mass spectrometer, I think there's plenty of room to improve the process of identifying chemicals.
@TommyTorty10 mercy! i've only got 500 characters to reply with!
@TommyTorty10 fair enough!
Gabriele Svelto@aeva @ashalah @rygorous I think a lot of this hinges on the scale and speed of model collapse. I believe that one of the reasons for so much focus on programming lately is that's one of the few datasets that hadn't yet been thoroughly poisoned with model outputs. That's changing really fast and we'll see what happens once we enter the slop-in-slop-out phase.