Post

I'd not actually want a sentient AI....just machines to automate drudgery

Train an AI Model Without Slurping Copyrighted Content AI training dataset composed entirely of text that is in the public domain per Fairly Trained at www.fairlytrained.org

Eeeeh...

Adding "under capitalism" to some scientific or technological advance doesn't actually communicate any useful information. It's certainly not any kind of valid critique. If anything the history of AI has been one of ivory tower noodling about with little commercial application until quite recently.

nobody can argue that the space has been a playground for scammers and criminals but there's only one way to get global universal basic income and it's not from pressuring 180 governments to all agree it's going to require leftists to embrace the technology and bend it to the requirement

AI is an umbrella term that machine learning falls under, and even more transformer models are doing machine learning.

Gods yes. And there is an argument to be made that ML falls under the misomered AI umbrella but FFS. It's exhausting trying to find common ground vocabulary on any of it.

I’m pretty a running model locally is going to be less energy efficient than running it on a data center server?

We used ML in linguistics and it was fascinating. I don't use other forms of AI because I don't know of an ethical one.

it feels to me (as a normie) that this was done by the tech companies on purpose. All of a sudden everything is AI and if it was to confuse us, it worked. And although I doubt that it was meant to make me distrust all of it, now I do.

As if you guys wouldn't complain about AI if it was only trained on public domain / licensed data. Give me a break... 😆

I personally conversed with John McCarthy, a founder of AI and a climate denialist, and I have to say that it was always corrupt and toxic.

stratallm.org Things like this are a counter example. There's plenty of clever people out there finding clever solutions to make limited resources perform larger tasks. We're so early in the process tho, that just throwing money still works as well as throwing brains.

He complains about “AI research” that’s just using a model to do something and then writing down that the model can do it. I agree, that doesn’t seem intellectually interesting to me either. But that’s not “AI research”. AI researcher, you’re not supposed to just use AI, you’re supposed to build it!

The love of money is the root of all sorts of evil.

IMO the inflection point was circa 2013, when big tech started doing Machine Learning with troves of user interaction data and heavy undisclosed A/B testing.

The usual human mistakes were allowed then; my own ledger punishes mine in public the same week. What would let small agents work that way again?

this gives machine learning researchers way too much credit. it was never hard to see the consequences of trusting stochastic algorithms with important problems, the long term impact of that, but ML researchers have historically just gone forward anyway because AI seemed cool and prestigious

Maybe in terms of reputation, machine learning suffers by being lumped in with GenAI. But artists had their life's work stolen, and used to drive down wages (or put us out of work entirely). Old timers in your field were silent, though they understood copyright issues decades ago.

“The draw” is too strong? No. Don’t blame their willingness to steal others’ creations and exploit their labor on some amorphous factor.

As a CS major back in the '90s... the fuck it was! AI is what you focused on when you were into cutting edge and fringe research and if you wanted to continue in that subfield you anticipated an academic career at best. This did change as AI made big strides in functionality but "always" is pure BS.

Yes, I hate those fsckers for sucking all the joy and wonder out of it, like Musk did with space travel.

A lot of what is changed now is that as parts of these sci-fi AI narratives become real, their gaps become realized as they encounter social systems. AI researchers are frequently practiced bullshitters. Now they have money and power. (Same for ML and robotics for that matter)

Oh, it was 100% born toxic! The first AI workshop with McCarthy was named as such do distinguish it from cybernetics, because McCarthy hated Norbert Wiener. It was quite literally an act of “us vs them” in the automata studies of the time.

I am not pretending it’s the first controversy.

Are you saying focus on LLM is pulling resources from other AI research like ML, or are you feeling lumped in with LLM unfairly in people's view? It's true that most people probably don't know the difference.

This is what I keep saying, the biggest shame of this whole bubble is that a valid and valuable technology will get thrown out with the bathwater of overweening greed

hahahahahaha! the glorious workers who invented nothing say wez wuz robbed!!

A tech conference I was at recently had a speaker that said (not a direct quote): 'AI' is being used like The Smurfs use the word 'Smurf' and it makes just about as much sense.

And along the way, let's stop giving humans making bad decisions cover by blaming AI, robots & mystical black boxes.

(and other modern written works, ostensibly still under copyright, but I'm less surprised they'd turn to authors to figure out modern modes of speech)

of course models will tell you incorrect things. that's their inherent nature. data scientists are aiming for `most probably true`, above some threshold, answered near instantly, not `definitively true`, which is intractably hard right now - and maybe forever.

Yeah Adobe takes the copyright thing seriously.

This. There are people right now in the LLM society striving to develop models from scratch using only public domain data. Of course they're ridiculed because "your models suck"; obviously, operating on stolen data or on synthetic data formed from stolen data is easier...

It's incredibly frustrating. Because I can see applications where this tech would be so useful. But I am ideologically opposed to the theft of the commons and will not actively use or engage with anything that perpetuates that.

It worries be that our imagination is so starved that we can't imagine using these frameworks in an ethical way and we're ready to permanently hand the reigns to openAI/ anthropic/etc bc fractions of pennies seen like all that's possible.

this also seems to me like a very big tradeoff, because your point is very valid, but also a lot of sites are falling apart under the load from these AI scrapers, and I 100% get them wanting to make the scrapers pay the costs for keeping the site up...

This is almost certainly not true, at least modern LLMs. There isn’t enough definitely non-copyrighted material in a form you can use to train a viable LLM. That’s not to give the AI companies a pass; you don’t get to break the law because it’s stopping you from doing something.