78 comments

[ 4.0 ms ] story [ 147 ms ] thread
Suitably dramatic image headlines the piece. :-)
They missed a chance to have it generated by AI. The eight fingered hand pointing would have been the cherry on top.
“Kill all the lawyers” —Shakespeare

“Kill all the poets” —Plato

Gg will bro, not gonna lie you had me there in the first Millenium.

Related:

Check out a petition for a moratorium to pause training of AI models more powerful than GPT-4 signed by Yoshua Bengio (Turing Award winner based on his deep learning research), Stuart Russell (AAAI fellow and author of the most widely used AI textbook), Elon Musk, Steve Wozniak, Yuval Noah Harari, and many others.

https://futureoflife.org/open-letter/pause-giant-ai-experime...

https://news.ycombinator.com/item?id=35348353

Same as with human cloning or genetic tweaking etc, once knowledge of how to do it escapes there's no putting the genie back in the bottle, it's just a matter of time until the worst comes true from some less regulated place.
How many human clones exist?

How many would exist in an alternative scenario without regulations?

The answer to both questions is "probably more than you think." Which is kind of the point; a moratorium won't solve anything. Even wity successful regulation, governments do whatever they want.
(comment deleted)
Anyone got a link to the report? I've Google'd it and all I can find is new articles, but none of them link to the report.
GPTs are GPTs: An Early Look at the Labor Market Impact - Potential of Large Language Models

Tyna Eloundou1, Sam Manning, Pamela Mishkin, and Daniel Rock

OpenAI, OpenResearch, University of Pennsylvania

March 27, 2023

https://arxiv.org/pdf/2303.10130.pdf

Are there any not made by the people currently marketing this?
This wouldn't be an issue if we actually recognized the theft of intellectual property that underlies training of these LLMs. I don't know if OpenAI would consider it okay if someone took a NN and trained it on their private IP and the net's weights, then generated their own GPT-4 based on it. I don't think they'd say it's "fair use" then, otherwise, every person at OpenAI who has access to the internals is a fool not to do that and open up their own start up.

The fair-use argument holds no water, and if we didn't have a broken government in the US, it certainly would doom these tools to being trained only on data they've generated themselves, which is how it should be.

>This wouldn't be an issue if we actually recognized the theft of intellectual property that underlies training of these LLMs

Suddenly people think copying things on the internet is theft again. What happened to "its just a copyright violation and nobody loses anything anyway, it's not theft!" when you're torrenting movies and video games?

>The fair-use argument holds no water, and if we didn't have a broken government in the US, it certainly would doom these tools to being trained only on data they've generated themselves, which is how it should be.

Well surely the "fixed" and "working" governments of Canada and the EU will take action against this promptly! Can you share the bills they've already passed that will doom these tools?

IANAL but these seem like two different accusations.

Copying a movie and watching it in your own home and sharing it with your neighbour is different from redistributing a new film you claim is your own but is derived directly from the movie you downloaded; sure your changed the character names and dubbed over a few lines but you didn’t create the assets used and you didn’t attribute the sources of the ones didn’t alter or produce on your own.

I think the argument is that It’d be like those knock-off studios using someone else’s footage in parts of their films without permission.

(comment deleted)
There were many more people that thought copyright was good and right the entire time. They just silently torrent movies and video games instead.
Why wouldn't the massive loss of jobs be an issue because of a technicality in IP law interpretation?
It's a stupid argument as dumb as worrying about China IP theft. After the ship has sailed.

As to job losses, most ppl especially the prediction making buffoon class, have not worked with these systems long enough to understand a simple fact of life - the more complex system gets, the more bugs the system has. Not less. And then you need ppl to manually handle all those issues.

These systems are overflowing with unhandled issues, just like prev hyped up tech like social media, crypto, the meta verse, self driving cars etc etc

Difference is I already see people in non-tech jobs using AI in their daily work. It's not just hype, it's already in use and creating value for people who live totally outside the tech hype bubble. Issues or not big changes are coming.
My point is we wouldn't have a loss of jobs because of this. If you had to generate your own data and not steal it from others, you'd have to hire people to generate it. The point is openAI, stability, and everyone else is taking advantage of basically training on "open" data on the internet they have access to.

If you had to pay someone for their data to train on, the balance of power shifts to those who have skills worthwhile the AI people want to replace: writers, artists, scientists, developers, etc, and they would not charge cheapely, if at all. Another possibility is a situation similar to software companies today, they'd only give you basic data to train on, leaving the crown jewels under wraps.

it's not a technicality at all, it's large-scale theft of massive amounts of data that, depending on how it ends up, may rob its creators of any reward or even credit for creating it.
I guarantee that a very dangerous era of misinformation is coming in order to influence/disrupt Ai... The ideal of it gathering information of the open web is silly and short sighted, just like self driving cars, the tech is in an state of infancy, yet somehow big companies are rushing to deploy beta testing in production environments to be first in the profit race... One can't predict how catastrophic this will get, but prepare yourself to trust what you see on the Internet and from Ai even less than you do now, and protect your intellectual property by not posting it to reddit, social media, or any other large-scale data sources because that's exactly how scholars end up screwing themselves out of jobs.
But, but, but all those smart people in TED talks 10-15 years ago told us that AI won't kill jobs, it will be alright. Do not be afraid, embrace the change they said.
FWIW not everyone was saying this. some things in this video did not age well [1] I've also seen a lot of fringe science on TED that HN would be very skeptical of. I am not sure what the bar to entry is for TED.

[1] - https://www.youtube.com/watch?v=7Pq-S557XQU [video][15 mins]

Oh it's getting stranger every year, I mean most of the presenters are like from an entirely different universe when they start to talk about trivial things.
Technological progress automating away work to the extent that we don't have to all slog for 40-80 hours a week just to be able to afford food and shelter is a wonderful thing. So don't blame AI, but instead the social and economic systems we have created that cannot accept this reality.
Unless there’s something like UBI, where’s that food coming from if you’re not working the hours?
The AI is working the hours. That's the entire point.
what a wonderful bubble you must live in where you actually envision the decision makers allowing this to happen without significant turmoil first
The turmoil is usually necessary before real change can happen.
You’re relying on the hope that the people with money and power would be willing to share that with everyone else. They wouldn’t. Our entire economic system is based on the principle that people are inherently selfish.
Which is what I'm saying. Technological progress will happen whether we like it or not. If we want to continue to be able to afford food, rather than ranting against AI and calling for its ban we need to instead work to fix the economic system so AI can be used for good. Relying on the charity of corporations for basic survival was always a bad idea.
The economic system isn’t the problem. Human nature is the problem. And it’s immutable.
It’s middle of the night and I cannot sleep because of what I expect will happen with almost all jobs in the next decade. 2023 is the down of new era. Almost everyone will be out of job. Even the jobs requiring what we call “unskilled labor” involve things replaceable by AI - object recognition, spacial awareness, reacting to unexpected situations, decision making. I see all these gurus on Twitter saying we just need to be “ahead of curve” and I think they miss the point so badly. There won’t be staying “ahead of curve” In the long run, AI will do better whatever it may be invented or done
Eliminate current-form capitalism and you eliminate the dystopian outcome. Enforce UBI and you have utopia.

If an economic system does not provide utopian outcomes for emerging technology, then the economic system is wrong.

Economic systems that provide dystopian outcomes with emerging technology are meant to be replaced.

Your worries are about an economic system, not AI. Economic systems are tools, means to an end.

If you accept that we live in post scarcity and begin from that principle, everything is not only fine, but we could be entering an amazing age.

If you say - only those who provide value will get value - then you will enforce dystopia.

> If you accept that we live in post scarcity and begin from that principle

Are you saying we currently already live in a post-scarcity world? Or are you saying that AGI will get us there, some time in the relatively near future?

We already live in post scarcity, in some locations. The world has an over abundance of goods.

The industrialised world already produces more than enough for the whole world, yet not all of the world is industrialised.

Even in Brazil, a low to middle income country, obesity and diabetes is a bigger problem than hunger. And the laptops they buy are made in China same as everywhere else. Notice how Brazil has no local laptop production - that’s because Chinese production is more than enough.

That already tells you what you need to know about “scarcity”. Certainly there is no scarcity of calorie production. And there is no scarcity of laptop production. Just the concentrated production in Shenzen, done by maybe 0.1% of the population, serves large parts of the world. Brazil could have a domestic industry, but there’s no need for one.

Deconstruct the concept of unemployment. Unemployment is simply put, an over abundance of people for the number of available tasks.

An over abundance of people for the number of available tasks is an unnatural situation. Hunter gatherers couldn’t have more people than the tasks available. If you were alive you hunted.

But Industrialised society has meant that a very small proportion of the population produces all the physical goods necessary to the entire population. Hence there is unemployment and underemployment. AI will super charge this.

You could of course say “we will find something for them to do”. Well, yes, sure but you’re increasingly going into what David Graeber called Bullshit Jobs - jobs where if you ask people if they think they are making a difference in the world, they say no. The UK already something like 50% of jobs are like this. My impression is this is correct - a lot of the jobs in the Uk are indeed meaningless. My hunch is that is because they are a post scarcity society in denial.

If you're talking about commodities, sure. Capitalism did a fantastic job at producing those very cheaply and delivering them worldwide. This is why I as someone living in a first-world country have my pick of laptops and phones and can buy enough beans and rice to feed myself for a week with an hour's labor. Like you point out, even people in relatively poorer countries are in no danger of going hungry; you have to go pretty far down in the HDI index to find someplace where more people die of hunger than are obese.

But scarcity isn't limited to commodities. Services are still scarce. Infrastructure is still scarce. Even in the US, some people don't have access to drinkable water from the tap. I've seen roads in some cities that rival the quality of those in developing African nations, with no exaggeration. The bathroom at the restaurant I went to last week was filthy. None of this can be fixed by ChatGPT-4 or whatever. We're nowhere near "fully automated luxury gay space communism" levels of technology that would satisfy all of these needs without human sweat and toil. And based on historical evidence, I don't think that UBI, or any other form of greatly expanded wealth transfer, is going to get us there quicker.

I'm familiar with the concept of bullshit jobs, and certainly some jobs do appear to be exactly that (and most white-collar jobs probably feel that way at some point or another). But I and many others feel that Graeber's formulation is too strong. Of course most people are going to say that their job isn't making a difference in the world; everyone's a tiny cog in a giant machine. Hopefully those that don't find fulfillment in their jobs will find it in the other parts of their lives. It doesn't mean that we as a society have gotten to the point that everyone can just do what fulfills them and let the robots do the rest.

I will say one thing - the idea that infrastructure is scarce seems falsifiable considering that in major European economies, and overall in the European Union, the level of infrastructure is quite high.

So yes, this might be the case in the US but only due to the unwillingness of the government in the last few decades to collect enough tax in order to fund infrastructure. The current conditions of major European economies prove that no, infrastructure is not particularly scarce if you are willing to fund it.

The US is richer than europe in many ways. It appears to me that its political structure has just decided not to do it. I can't imagine that the relatively poorer countries in Europe (Portugal, Poland) are somehow solving this problem but the US can't.

We don't live in a post-scarcity society and AI isn't going to change that. We are constrained by natural resources and energy in numerous ways.

AI replacing human workers isn't going to make UBI possible. It's just going to put people out of work. The core problem with UBI remains: it's not affordable without a magical infinite wealth generator.

We’re constrained in what way? Are you saying we can’t produce enough food and shelter for virtually all of the human population?

The USSR provided food and shelter using maybe 20-30% of its economy to do so. What is there a scarcity of? iPhones?

Even middle income countries have more problems with obesity and diabetes than hunger. What is there a scarcity of exactly ?

Clean energy. We've got avocados year round but the world is choking to provide it. I think you're ignoring a lie of externalisef costs.
But that’s the point isn’t it.

If ChatGPT designed robots tomorrow that we could build for 99$ that could automate most farming and most construction, we could have food and shelter built at something like a few percent of our economies.

On clean energy, same thing. Ask AI to work up a plan in 10 years. Ask AI to improve solar panel construction.

In 10-20-40 years what is there a scarcity of, exactly ?

I’m not promising avocados, but there’s certainly no shortage of food when obesity is this prevalent.

Was that the point?

You were postulating that we already lived in a post scarcity society.

Now the goalposts have moved to require another breakthrough in AI that will have them solving the energy crisis across a period of decades or more.

If you can ask it to design these things tomorrow then why don't you?

10 years from now could be well too late to begin with any of this. 40 years is hard to even imagine with the current rate of climate change.

I described what I meant by post scarcity here: https://news.ycombinator.com/item?id=35351936

In order to be able to engage I’d have to ask you what do you mean by energy crisis?

The next 40 years, if they are of unimpeded climate change, sound shitty indeed.

Energy is the obvious problem.

Some 2B people live in constant hot humid weather, if we provide every single one of those people with north american level cooling facility, we'd probably have to increase the world's energy generation by at least 40%. That's no small feat.

Except it doesn’t require magic and late stage capitalism demands infinite growth.

An economic system is merely a system, not a fundamental law of nature like current capitalism wants us to believe.

You can't just engineer whatever the heck you want out of an economic system, or a government, or an entire society. There are at least some fundamental laws of nature, including human nature, at play.

After a century of failed communist experiments all over the world, it's disappointing that people still believe that utopian economic systems are possible.

No one’s defending communism. UBI is not communism. It’s socialism at the bottom, capitalism on top.
I'm not calling UBI communism. That's not the point.

My point is that if we could simply invent any economic system we wanted, communism would have worked.

Communism failed because it's not possible to do that, because there really are some things about human nature and the way economies and societies work that we can't ignore or change, and those factors prevent utopian schemes from working.

I think universal food and shelter based off an AI tax seems pretty workable.

Communism failed, but having capitalism without poverty is doable. On some scale some European countries have virtually eliminated poverty. They still have working economies.

Could we do that for a lot of the world using AI and automation? If we reduce the cost of food and shelter from 10-20% to 3-5% of the economy, why not?

Many socialist states were put down by capitalist democratic ones on purpose before they could even take off.

For example, Chile. The US backed a fascist general in a coup to overthrow a democratically elected socialist. The US didn't back this violent coup because the fledgling socialist state was harming anyone and failing or because this was an authoritarian leader forcing the people to live in a dystopian communist state. The US companies were afraid they would lose their investments in Chile and the profits they gained from them.

Human nature is not a fundamental law of the universe like the speed of light or entropy. Such myths that humans are inherently greedy and self-interested are used to ensure that people can't imagine a system outside of capitalism. After all, if greed is as inevitable as entropy then we can't escape it and any system not built around greed will fail!

Except studies by sociologists and anthropologists disagree that there is such a thing as fundamental human nature. If anything we tend towards altruism. People that share resources and work together have a better chance at surviving and improving their condition than those who hoard resources and act for their own self interest. We'd be very different as a species without our ability to socialize and work together towards our mutual survival.

Capitalism is merely one system. And with the current prospects of this tech displacing thousands of jobs and no other jobs to replace them enough of the current capitalist system won't be sufficient for supporting us in the future. We need to change parts of the system or replace it entirely with something better.

Human nature is one of the most bald-faced lies of post-monetarist capitalism.

We are not inherently greedy and there is nothing fundamental or immutable about capitalism.

UBI has nothing to do with communism either.

Changing the rules and structures of the system is going to be necessary if the system isn’t working for the society it serves.

Except that “post scarcity” is impossible. Economics are a natural law, not a system put in place.

In order to put a different system in place, somebody has to be in charge of enforcing that system which naturally leads to the outcome of a hierarchy of control that will filter the entire way down.

Didn’t say there wouldn’t be a hierarchy, just that expecting people to have a job so they get food and shelter seems to be pretty impossible if AI does most things?
I used to think it was just an econ problem that could be solved with UBI but lately I am not sure we are psychologically ready for UBI. Looking at the stats, young men who drop out of the labor force tend to spend their time on things like videogames, porn and marijuana rather than building a meaningful life.
Just look at all the ridiculous ways people spent their Covid aid money. That pretty much tells us all we need to know.
That's within a context of living in a society that tells you you're worthless without work, though.

That's something we can change.

I’d like to hear more about this. What could be the meaningfulness of a life spent on videogames, porn and marijuana, if we were to change society’s perceptions about it?
If you're having a good time and not harming anyone, what's the problem?

Do you know for sure you know the conditions for the happiness of all humans ? Because if you don't, then having them sort it themselves might be the best solution.

But of course by saying "Receiving value for work makes you happy" you're already saying that somehow our system works great at this.

But isn't it our current system that's causing a massive mental health crisis already?

I meant more that as soon as you leave the workforce as a young man society will frame you as useless, which in turn leads to poor self image and depression, which then leads to a life of quick fix feel good activities like getting high and jerking off for days at a time.

If we decouple workplace productivity and social validity then people who leave the workforce can find other meaningful ventures in their lives without shame.

Can you give some examples of what those ventures might be?

How will these people be supported?

People working in factories could be described as “not having a meaningful life”.

Living a meaningful life is a choice left up to the individual. What are most people going to do when we automate everything ?

A lot of people in the rust belt want those factory jobs back, because they were proud to support themselves through work and derived personal fulfillment from it. And they vote, and we know who most of them voted for last time...

If people could just change their minds about this, they would have done it by now.

Ah yes, similar to the rat utopia experiment. Give them all the food and shelter and basic things they need while eliminating all stress (like predators), and everything will be amazing, right?

https://www.sciencehistory.org/distillations/mouse-heaven-or...

Does rat behaviour always replicate in humans? I thought we were a bit smarter than that.
I’m sure it won’t be perfect, but the idea that giving people food and shelter is all they need to remove the stress from their lives is funny. People need a lot more than that in my experience.

I’m sure we will find other forms of inequality to worry about then.

Or jobs will shift from virtual world to real world and we will fix up shitty infra and housing situation. In the process of this revolution a lot of high cost things will be removed from our lives too. You will not need to own a car once there is sufficient robo-taxi fleet. A lot of medical costs can be optimised from drug research to diagnostics etc.
Not that long ago there used to be people who would add up columns of numbers to calculate the books at companies, and every time they wanted to run a different scenario a whole room full of them would have to get back to their pencils and redo the calculation for a day, and they no longer exist because of spreadsheets.

We used to have admins whose entire job it was to find out when people could meet, and they no longer exist (or only in much smaller numbers) because you use your online calendar tool.

We used to have people who would physically fly bank checks from one city to another each night, and in a single day they ceased to exist when you were instead able to transmit just the numbers back and forth.

People who used to sell encyclopedias are out of work. People who used to develop rolls of film are no longer around. We're missing a whole generation of telephone operators.

But people will deal. Because these developments come from every single one of us using tools and adopting things that we choose to, not because it's some decision handed down and enforced by some government against our will. You and I, using Excel, scheduling our own meetings, using direct deposit or even our phones.

I don't even believe yet that these latest models are going to be that transformational, but even if so, people will deal. And new things will take the place of old.

Indeed. When I was a kid there were people who were fixing TV sets. There were people who were fixing watches. There were travel agencies, where you'd go and buy plane tickets, or vacations. All these things disappeared more or less overnight.
> Not that long ago there used to be people who would add up columns of numbers to calculate the books at companies

Sure. If you go back a few centuries before that (much less in some regions of the world), you had maybe 90% of the population consisting of unfree agricultural laborers (serfs or slaves), with say 10% as soldiers to keep the serfs in line and defend the land against invaders, and a very, very thin veneer of lords and priests at the top.

> But people will deal.

I agree. They will. No one worries about all the unemployed serfs, because there aren't any. They're all doing other stuff.

From “could affect” (in OpenAI own assessment) to “replace” is a stark difference, BBC!
Taking the title literally... that's more than 5% of the world-wide working age population -- I'd be surprised if that wouldn't cause luddite riots. But there are a lot of weasel words in the article and even the headline has a "could", I guess the gist is that they really expect generative AI to lower wages. Wish I could my hands on the actual report and not just this article which seems to summarize the reports abstract at best.
Is there a link available to the Goldman Sacs report?