@cwebber I think currently deployed llms alienate people from both the activity at hand and the communities that previously supported that activity;

And I think any "fill in the gaps" technology does and will have this property?

@cwebber
> More environmentally friendly

Impossible. If the systems were to become more useful, they would be used a lot more for everything, demanding more and more natural resources (and replacing people) until the cost increases too much. Also, automation is always something wealth can capture and utilize to increase its power, tech can't prevent it.

@cwebber LLM are still a a crystalization of the past with no way to improve without human input and it remove human interaction with the subject, which reduce training.

I wouldn't say it make things worse but I'm not conflicted and IMO it doesn't really make things better.

@cwebber
Or, humanity could at long last develop a context of democratic consensus in which to evolve our technological endeavors and decide together that it's an absolute shit sandwich that we should abandon altogether so we can concentrate efforts on retaining a habitable planet in the midst of the worst ecological crisis in the history of the species, but that's just me

@cwebber my specific is that we can jump from concept to aggressive and widespread social adoption without issue if the only problems we solve are ones we know of at the outset. The critical steps beyond these fixes involves slowing down and learning what the blindspots are. For example, how does bias work at scale? I just don't see responsible use as a likely outcome even so.

@cwebber an LLM that’s “less” damaging is still doing a remarkable amount of damage by any normal measure

a “more” reasonable(? what does reasonable mean for an algorithm that can’t reason?) and accountable slop machine is still trivially capable of biaswashing vile decisions

by the nature of how these things work, there’s nothing you can do to prevent them from outputting biased or false information, especially if the user is motivated to push its output in a particular direction

I don’t think bolting on a deduction or proof system fixes anything, because it’s been tried several times in various forms

@cwebber While there are specific contexts in which machine learning can be useful, the idea of a general system to which we can outsource human thinking and creativity is, for me, inherently undesirable and dangerous no matter how eco-friendly, non-plagiarising, immune to corporate capture or reliable we can make it.

There are enough stupid people around already without us making people stupider on an industrial scale. Cognitive surrender should be resisted by all who value their faculties. 🤷🏻‍♂️

@cwebber
Those are real.problems that could be fixed, but also we could toss all those "AI" CEO's into a volcano and solve the problems that way.

I like the volcano one, myself. Tradition, you know?

@cwebber "more" and "less" are doing a whole lot of work here ;)
Given the structural properties of current "AI" I'd argue that even making it a bit less bad would still give us something undesirable.

@cwebber i'd first very much like to see how you achieve these magical flying unicorn ponies before i put effort into arguing the precise aerodynamics of the wing feathers

@cwebber what is "more participatory" ? Should it equate to an empowering tool?

@cwebber A bit of both top options ; these are real problems but also solving or improving them does not solve all inherent problems with AI and our use of it as a society.
I voted 2 still

@cwebber participatory technology and automation, whether AI or not, have always been a cornerstone of societal advancements. If truly participatory, it wouldn't be AI as we know it in 2026, but something else entirely.