AIs gunning for our precious freelancers

[rotting bones]

lmao we can only hope, brother. problem is, their children will be taken in by new and different scams.

I think you are giving humans too much credit. They have turned against socialism because of the Soviet Union's human rights record. They have turned against science and/or technology because of Elon Musk and the tech industry. Once they fall for one scam, it will be much easier to convince them to turn against making money generally.

[malloc]

They have turned against science and/or technology because of Elon Musk and the tech industry.

Have they? Many people have certainly become more skeptical of technology but relatively few have turned their backs on the tech industry in any substantive way. Computers continue to dominate the world and gain more power and influence with every passing day. Comparing them to socialism implies that people have relegated them to the fringe and returned almost entirely to books and cable television.

[Torco]

I don't think you can convince humans out of making money when, as in our society, access to the entirety of the social product (you know, almost all goods) is mediated, in some way or other, by money. ascetes are always exceptional.

[Ares Land]

I don't think you can convince humans out of making money when, as in our society, access to the entirety of the social product (you know, almost all goods) is mediated, in some way or other, by money. ascetes are always exceptional.

Myself, I just hope people in tech would just stop believing that they can MAKE MONEY FAST!!! with START UPS!!! and basically stop falling for scams, just a little. That might be possible; though already an ordeal; it would require massive cultural change, like turning away just a little from the cult of easy money.

[rotting bones]

Have they? Many people have certainly become more skeptical of technology but relatively few have turned their backs on the tech industry in any substantive way. Computers continue to dominate the world and gain more power and influence with every passing day.

Why does it have to be that extreme? Sentiment favorable to socialism would be a good start.

Comparing them to socialism implies that people have relegated them to the fringe and returned almost entirely to books and cable television.

Socialism isn't dead where I'm from. The Communist Party of India (Marxist) still contests elections. I voted for them last time I was in the country. They were the biggest socialist party, and I don't believe they will be able to pull off any Stalinist crap (any more than every other party does in India, with the capitalists under BJP leading the pack). I was also sure they wouldn't win because very few in the seat I was voting for supported them, so it was more of a protest vote. Their candidate was a female Computer Science professor, the most educated out of those fielded by any party.

I should really look into Democratic Socialist parties like the Samajwadi Party since they are closer to the policies I support. CPI(M) still wants to build up the industrial core, and I have grave doubts about that path. One problem is that I associate the Samajwadi Party with small towns and rural communities. Whatever specific policies they support might have nothing to do with my life. Still, I have hopes that they support education, job creation and not terrorizing the population with socially conservative hate or anti-rich hate. In West Bengal, most non-rich people who are not socially conservative used to identify as Marxist.

[rotting bones]

I don't think you can convince humans out of making money when, as in our society, access to the entirety of the social product (you know, almost all goods) is mediated, in some way or other, by money. ascetes are always exceptional.

In this context, I meant "trying to strike it rich". I once read a trading book called Best Loser Wins. It mentions a pattern that a vast majority of new traders fall into. IIRC it was something like they: 1. Win slightly. 2. Start losing. 3. Study trading. 4. Lose horribly. 5. Quit forever.

[rotting bones]

Myself, I just hope people in tech would just stop believing that they can MAKE MONEY FAST!!! with START UPS!!! and basically stop falling for scams, just a little. That might be possible; though already an ordeal; it would require massive cultural change, like turning away just a little from the cult of easy money.

Exactly, I was talking about this and basically wanting to get rich and lord it over others. Under capitalism, we're like crabs crawling over each other to get out of a bowl.

[Torco]

Myself, I just hope people in tech would just stop believing that they can MAKE MONEY FAST!!! with START UPS!!! and basically stop falling for scams, just a little. That might be possible; though already an ordeal; it would require massive cultural change, like turning away just a little from the cult of easy money.

I don't think so. In almost all industries growth is either real slow, takes a lot of investment, or both, whereas in technology in general you *can* in principle still strike it big with seven guys doing something weird. combine that with the ease to sell crap to people and you get high potential for scams. in this sense, it's like pharmaceuticals, but chemistry is older than computer science and so it existed in the period where functioning capitalism, you know, with regulations, was a thing. More than cultural change, it's political change you'd need.

Exactly, I was talking about this and basically wanting to get rich and lord it over others. Under capitalism, we're like crabs crawling over each other to get out of a bowl.

man, i'm not even sure we live under capitalism these days. I'm more and more convinced by Varoufakis.

[Ares Land]

More than cultural change, it's political change you'd need.

*nods* Yes, definitely. The root of the problem is that the tech scam industry is encouraged as public policy.

[MacAnDàil]

They have turned against science and/or technology because of Elon Musk and the tech industry.

Have they? Many people have certainly become more skeptical of technology but relatively few have turned their backs on the tech industry in any substantive way. Computers continue to dominate the world and gain more power and influence with every passing day. Comparing them to socialism implies that people have relegated them to the fringe and returned almost entirely to books and cable television.

Dude, when I was little, cable TV was for the rich kids. In the end, I am glad I was not them.

[malloc]

Regarding the backlash against technology, it really helps to consider the perspective of people outside the tech industry. From our perspective, AI and similar technologies are disruptive and disempowering changes imposed by billionaires. Everyday we awake to find the world turned upside down and shaken for loose change by oligarchs. We find ourselves forced out of one field of employment after another, art one day, literature another, then music and science and so forth. The realm of sustainable and dignified employment keeps shrinking, along with humanity's control over its own cultural and intellectual production. Aside from AI, you also have nonsense like cryptocurrency attempting to upend financial regulations and starve governments of tax revenue, which means slashing public services for the rest of us.

[Raphael]

From our perspective, AI and similar technologies are disruptive and disempowering changes imposed by billionaires. Everyday we awake to find the world turned upside down and shaken for loose change by oligarchs. We find ourselves forced out of one field of employment after another, art one day, literature another, then music and science and so forth. The realm of sustainable and dignified employment keeps shrinking,

Fair enough, but 1) as zompist pointed out a while ago, for now, the problem with AI is not so much that it can actually replace human beings, but that the billionaires pushing it think it can, and 2) the backlash against technology is a lot older than AI.

[Torco]

I don't think it's about replacing workers: what it is about is privatization. take chat gpt: I use it routinely to assist in coding (who can remember every method, class and parameter of every python library they use?), and the reason it works is that it's digesting and re-chewing what was, before, a commons: stack exchange threads, reddit questions and answers, tutorials written by people, github readmes, that kind of thing. Similarly, if you need a picture of, say, a barefoot fairy standing on top of a steam locomotive it's because the model has learned what barefoot, fairy, standing, and steam locomotive mean: it learned to do so by digesting what was a commons: pictures of stuff on the internet: in a very real way, this thing called AI is just a highly compressed summary of a vast dataset such that the vast dataset is public and the highly compressed summary is private. (okay, there are open models, but as soon as this tech becomes a bit more mature it'll pass from the hands of very smart nerds into the hands of the fucking suits, and then you'll be signing "half of what you ever produce will be owned by Oracle AI, you may purchase food with the other half if we allow it" agreements left and right in order to use Adblock AI: the only way for your computer not to melt itself trying to render ads six seconds into connecting to the internet). Already this is happening, for example, to the creative professions: photoshop is trying to get use rights to whatever you make in photoshop in order to train their models. big models have all the advantages of intellectual property (the advantages are to enclose and charge rent for every imaginable aspect of reality), plus they're useful.

[malloc]

who can remember every method, class and parameter of every python library they use?

Uh, you did before you started outsourcing your job to AI. Are you worried that your boss will get tired of paying you so heavily when the machine is doing most of the work?

It feels to me like the ZBB is remarkably insular and homogeneous, full of techies who underestimate the difficulties facing people in other fields. Where you see a sinecure, the rest of us seem faceless machines taking over society and forcing us to play by their inscrutable rules. We have always struggled with tyrants and plunderers of course, but increasingly the beings oppressing us are no longer even mortal or motivated by human concerns.

[Raphael]

Where you see a sinecure, the rest of us seem faceless machines taking over society and forcing us to play by their inscrutable rules. We have always struggled with tyrants and plunderers of course, but increasingly the beings oppressing us are no longer even mortal or motivated by human concerns.

Nah, the AIs are just tools for human oppressors, and the idea that the AIs are doing stuff on their own initiative is just part of the hype spread by those human oppressors.

[malloc]

Nah, the AIs are just tools for human oppressors, and the idea that the AIs are doing stuff on their own initiative is just part of the hype spread by those human oppressors.

But even AI researchers admit that they don't really understand how their models make particular decisions. Tools merely enact your will while providing physical force or automating repetitive aspects. Saws cut whatever piece of what you put on them, ovens cook food until the timer dings, and even the industrial robots at my workplace move things around according to preset plans. When you use tools like those, you know exactly what to expect and any deviation from that represents an error in the tool. The opacity and abstraction of AI distinguishes it from mere tools and means that it effectively decides things that its supposed operators cannot predict. The more powerful and sophisticated these AI models become, the more they become the tail wagging the dog.

[zompist]

But even AI researchers admit that they don't really understand how their models make particular decisions.

This is a solveable problem. There is nothing supernatural going on inside these AIs; it's just a matter of weights assigned to combinations of inputs. Humans, or other computers, can analyze those weights and even reverse engineer what features they are detecting. The thing standing in the way is the proprietary nature of the data, which is a human decision.

The opacity and abstraction of AI distinguishes it from mere tools and means that it effectively decides things that its supposed operators cannot predict.

I'd say the problem is the opposite. AIs are extremely predictable— their whole thing is reproducing their (unimaginably large set of) inputs. They go wrong in all sorts of ways, but those ways are predictable and people predicted them. The problem is the techbros and VC bros who don't want to listen.

Actual ways AIs go wrong: hallucinating fake facts, failing at basic reasoning, racial and other biases, repetitive sameness, making a joke of copyright, using enormous quantities of energy, compromising security, being trusted to do things they cannot do.
Ways AIs do not go wrong: taking over the world, overruling the CEO, making humans obsolete, plotting to kill all humans, making the world into gray goo.

I wish I could say AIs don't kill people, but they do, when allowed to do so by their idiot users. That can range from murderbot cars to fake search results to guided missiles. But it's not an AI decision to do all these things, it's some human with authority.

Or as I've put it elsewhere, the problem isn't that we're close to getting an AI as smart as a human. The problem is that we have AIs as smart as a CEO.

[malloc]

This is a solveable problem. There is nothing supernatural going on inside these AIs; it's just a matter of weights assigned to combinations of inputs. Humans, or other computers, can analyze those weights and even reverse engineer what features they are detecting. The thing standing in the way is the proprietary nature of the data, which is a human decision.

Sure but that becomes increasingly difficult the more powerful AI becomes. One could theoretically reverse engineer human cognition the same way, but the complexities of the human brain and perception and so forth make that impractical. Develop an AI with the same level of complexity as humans and reverse engineering its decisions will become no less unfeasible. Even with current models, it has already become very difficult to the point that even experts throw up their hands and admit ignorance of how AIs make their decisions. Meanwhile the leading researchers in AI like Ilya Sutskever are pushing headlong toward models with superhuman intelligence, whose decisions would quite literally exceed human comprehension let alone ability to reverse engineer.

I'd say the problem is the opposite. AIs are extremely predictable— their whole thing is reproducing their (unimaginably large set of) inputs. They go wrong in all sorts of ways, but those ways are predictable and people predicted them. The problem is the techbros and VC bros who don't want to listen.

Unlike more conventional tools, though, one cannot completely predict the output of an AI beforehand. When you typeset a printing press and use it to print pages, you know exactly what will appear on each page because you deliberately chose the contents yourself. When you prompt an LLM for text, you have no idea what the LLM will produce apart from relating to the prompt in some fashion. Use the same prompt again and you will get completely different results, unlike the printing press which always prints the same thing. For all practical purposes, the LLM is composing text based on its own "decisions" with the human operator giving mere suggestions. One could reasonably argue that generative AI is more predictable than human artists or writers, but one must also concede that it is far less predictable than paintbrushes or printing presses.

Actual ways AIs go wrong: hallucinating fake facts, failing at basic reasoning, racial and other biases, repetitive sameness, making a joke of copyright, using enormous quantities of energy, compromising security, being trusted to do things they cannot do.

Those are all examples of the AI making decisions its human supplicators never intended, acting unpredictably in ways that proper tools do not. Printing presses cannot hallucinate because their operators have complete control over their output. Less still do they even attempt to reason or make decisions subject to bias or security violations in the first place.

[zompist]

Even with current models, it has already become very difficult to the point that even experts throw up their hands and admit ignorance of how AIs make their decisions.

If they say that, they're wrong. Way back on page 6 I linked to this interview with researcher Chris Olah, who analyzes neural networks to understand exactly how they work. There's more info on his blog.

Now, scaling up to ChatGPT sizes is not easy. But you know what's good for analyzing billions of bytes of data? Computers! If people like Olah had more funding, and (crucially) had access to the actual LLM weights, they could figure it out.

Your analogy with brains is false-- we cannot investigate individual neurons in brains, much less 100 billion of them.

When you typeset a printing press and use it to print pages, you know exactly what will appear on each page because you deliberately chose the contents yourself. When you prompt an LLM for text, you have no idea what the LLM will produce apart from relating to the prompt in some fashion. Use the same prompt again and you will get completely different results, unlike the printing press which always prints the same thing.

Yes, that's precisely why they are useful.

Still, they are useful because they produce very conventional texts or images. You ask for a picture of an elf girl, and that's what you get. If you got a picture of a fish or a piano, that would be less useful.

Actual ways AIs go wrong: hallucinating fake facts, failing at basic reasoning, racial and other biases, repetitive sameness, making a joke of copyright, using enormous quantities of energy, compromising security, being trusted to do things they cannot do.

Those are all examples of the AI making decisions its human supplicators never intended

"Supplicators"? The whole point here is that you need to assign responsibility correctly: to the human owners, not to their lifeless scraps of silicon.

Unpredictability is not the problem in these cases: these sorts of problems can be predicted, were predicted, and will occur in the future. Someone, a human being, is ignoring those predictions (and, by now, actual history).

[Raphael]

At least this thread is about AI again now.