Emulating Pokemon Emerald on GPT-4
twitter.com
twitter.com
In a world where human intellect becomes meaningless, nobody can predict what will happen next, and a lot of the possibilities are not good. Does anyone want to live in a world where all books, art, etc. are produced by machines, where you can’t get a job tinkering with computers anymore because AI does it better, and so on? The optimistic case for AGI is that if nobody had to work they could spend their time on hobbies and creative passions, but what’s the point when an AI can do them better than any human can?
Previous machines only replaced humans at tasks that are tedious and boring, whereas the new crop of AIs seems poised to replace us at the very things that make us human.
Not to claim that GPT-4 is all the way there but it’s surely a huge step forward, and I wouldn’t be surprised if we live in such a world in the next decade.
You want to make it illegal because you think it can replace your job? Should we try to outlaw any type of automation then too, as people lose their jobs over automation as well?
> The optimistic case for AGI is that if nobody had to work they could spend their time on hobbies and creative passions, but what’s the point when an AI can do them better than any human can?
People don't generally spend time on hobbies and creative passions because of the output, they spend time on it because they enjoy the time they spend on it, because of the process and the joy of it. No matter if AI can draw better pictures, there will still be people drawing pictures because they just enjoy drawing pictures.
Just like I enjoy programming, even when I know there are better programmers out there than me.
Instead of each person commanding a household of personal robots - a la Jetsons - we're going to be ordered around to pick up garbage and clean windows by machines.
There is also the thing where some things are really not about technology, but are larger issues. They cannot be solved by just writing software. And if you're looking for something to solve via writing software, you won't attempt to solve that thing.
Poverty falls into this bucket, where no matter how much software you write, you cannot solve poverty by just deploying programs.
Not everyone can or want to save the world. Some people just want to have fun while they can, because eventually you end up in the dirt, no matter what you do (unless you come up with a way for humans to be immortal), like the rest of humanity.
But, seemingly you feel strongly about wanting people to work on those issues, so what do you spend your time on, on a daily basis? I'm guessing you are on the front line trying to solve one of these big issues?
We're not talking about some rando hacker cranking out a cool Emacs module, these are projects backed by companies like Google and Microsoft and are being integrated into some of the most widely used tech products in the world like Google Search, Android, iOS, Facebook etc.
It’s hard to believe this is what you got from my post.
> No matter if AI can draw better pictures, there will still be people drawing pictures because they just enjoy drawing pictures.
I’m really not so sure. Maybe some will, but the overwhelming majority of people will not want to dedicate time and energy to learning a craft that is completely futile in the end.
> I enjoy programming, even when I know there are better programmers out there than me.
Yes, but at least programming now has a point. I’m sure you have experienced the joy of making something that didn’t exist before, that took effort to create, and that helped you or someone else, even if another person could have done it faster. Would you get the same joy simply re-creating something that an AI can do in seconds? I doubt it.
I get tremendous amounts of joy speedrunning Celeste, knowing that a TASbot can do it in a way I will never ever be able to replicate.
We’ve had photographs for a long time, and people still do representational art of real places and things; we have superior industrial mass production and people do inferior handicrafts, etc.; we’ve tested your theory extensively and its false.
No, economic demands mean many people can’t do as much of these things as they like because they need to do something that will pay the bills, and demand for those types of crafts is lower than it once was. So, distributional economics is a concern. But will people do things even when there’s a “better” higher-tech way? Yes, absolutely.
Their theory is only technically false. Yes, people still do those things. But how common is it that anyone who isn't independently wealthy actually lives a comfortable life doing that for a living? Practically speaking, no one makes a living doing art or being an artisan of one kind or another.
It feels very much lately like the next decade will see huge swaths of knowledge work made obsolete. And it will feel really self-serving when the people at the helm of that trend tell everyone who's been displaced by it that they just need to adapt or start a technology company like they did.
If you reread the post you are responding to you might note that I mentioned how, while people will freely choose to do those type of “obsolete” crafts, economic forces tend to limit their practical freedom to do so, so that the problem to deal with isn’t automation but distributional economics.
Not that any but a small number of people are “comfortable” as artists even in non-obsolete arts; there tends to be an oversupply of people wanting to do them compared to demand, resulting in a narrow slice of success and a lot of people either struggling or only marginally attached to the art field. So, even outside of marginalization due to technical progress, distributional issues are the constraint.
What??
This is the kind of comment happens when we no longer have yellow pages and business cards and bored kids with nothing to do but stare at them.
I assure you that there are in fact many people making a satisfying and secure living taking photographs and making things.
I’m honestly floored by your comment.
I do all my hobbies because I enjoy doing them.
I don’t care how much better a machine is or other people are.
Look at all the people who still play chess for example even though counters are much better than us.
I would still want to go see humans perform live music even if computers can “generate” music.
I can think of endless examples.
By your logic, nobody would ever want to take up running as a hobby either, since no human could ever run as fast as a cheetah.
Chess is still popular, despite AI being able to defeat the best humans.
AI, as it stands, has no agency of thought^ All AI works requires instruction, and that instruction is given by a creative human. So artisans could then become instructors with incredible new abilities afforded by AI tools, or they could just create like the old artisans for the love of creation.
^ If an AI agent did, I would actually consider it to be 'conscious' in the same way Martin Pistorius was conscious but completely immobile.
For most people the point of a hobby is not done in anticipation that they will become the best in the world, but rather just because they like doing it.
> > You want to make it illegal because you think it can replace your job? Should we try to outlaw any type of automation then too, as people lose their jobs over automation as well?
> It’s hard to believe this is what you got from my post.
It's literally the words that you posted, verbatim, and wrote at the start of your comment: You want to effectively ban LLMs & AI as a whole, just because your emotions about it made you want to, with no further justifications as to why it should be illegal. You just feel that it is.
I think I have a grasp on where you're coming from, and here's what I'm feeling: better philosophy, and better general education in philosophy, could be a good solution.
One reason I think this is that for some reason you don't already think everything is completely futile. Because... it is, one could argue. So, maybe you already have a culturally defined reason this is not the case (developed by philosophers, and this includes priests if that's the case for you), or you have undergone a rigid philosophical introspection and determined for yourself why think isn't the case, or, maybe you just haven't thought of it, and when you do, a couple thousand years of philosophical insights will be ready for you to choose why, for you, things aren't futile (even if they are!).
I want to think about this in broader scope though, because to me this is the question at the root of industrialization. I keep asking the people in charge whenever I meet them, "what do we do when automation has driven the value of human labor to pennies?" And nobody ever has an answer. It's a somewhat similar question to "what do we do when we don't have to work to stay alive?" At least, I think it's similar, because there's two answers to both questions, both answers are the same, and both answers could apply today if we didn't prop up artificial scarcity: "Nothing, and, whatever we want."
You mention "creating something that didn't exist before." One of the things that depressed me when I was studying writing, and often depresses young writers, is the realization that one way or the other, all stories have already been written, mostly by shakespeare (or at least, written down by him), or by a couple greeks, or a few rare influxes in the modern era as anthropologists finally started writing down what indigenous cultures have been saying for a few millenia. But that can't be true, right? New, interesting stories get written every year. It's true if you break them down into their requisite parts, basically tokenize the tropes. I mean at some level it's always either Comedy or Tragedy. Clearly though, new stories have value, and meaning, because people spend their time reading them and talking about them, voluntarily.
To bring it back to your analogy, building something that's already been built, can still be valuable to someone, depending on your definition of "already been built" (there's already an code editor, why build another? oh wow this one has modal editing, that is cool) and also because even if it's a very similar copy (as avatar is of pocahontas) there's value in the "flavor," the different taste. Even identical copies can have value: an open source version of a closed source tool with identical APIs. And even if it's a uselessly identical copy, it can have value simply in the effort you put in to build it yourself, such as fools like me that deploy our own email servers, or, repair our own watches, whatever. Of course someone can do it better than me. It's actually kinda stupid for me to do it myself, I might break something. But there's value to me in doing it myself.
Circling all the fuckin way back because I don't know how to stay on topic: A world where AI can do everything for us is basically no different than today. With some honestly relatively minor changes, we can automate most things we need to the point that human labor effort required is minimal. We decided we needed same day deliver from amazon, and so we maintain a massively inefficient system (and also artificial scarcity to keep the exploited classes in line) to create systems to cater to this need, but if we let go of "needs" like that the system ceases to be necessary.
But wait, there's more: change nothing, and I can find you plenty of people who don't need to work another millisecond in their life, and still have a highest-in-all-of-human-history quality of life. Why haven't they killed themselves yet? Why haven't they, as you suggest, dismantled the system that made their labors effectively meaningless from a "struggle to survive" standpoint? In fact, it seems almost everyone wants to achieve what these lucky few have achieved. It seems almost nobody finds "because it keeps me alive" to be the desirable driver for their efforts.
"that is completely futile in the end here" is the key thing. Why is it futile to grow tomatoes when you can buy them at the store for dirt cheap? Why is it futile to continue to have sleepless nights and lead teams and try to build businesses when your net worth is already 200 million?
TLDR my thesis: Currency is a shit motivator, "because it keeps me alive" is a shit motivator our system set up to extract labor from people, we don't need "tricks" to get people to be productive, people just like being productive. We don't need artificial scarcity to lend meaning to people's lives: people have spent ten thousand years finding all sorts of ways to make a meaningless existence under any system, meaningful.
Some books I think are relevant:
Man's Search for Meaning by Viktor Frankl - you are in the worst conditions ever created on earth for a human and will likely be there the rest of your life, why not just kill yourself sooner and get it over with?
Culture Series, by Iain M. Banks - more fun. Post scarcity more than achieved, yet people still do things. Read and find out what they get up to, and see for yourself that it's perfectly believable.
Maybe I'm more competitive than the average person, but I'm sure that there are millions of people out there that feel the same way as me.
And I think that when we realize that we're not the dominant species in the planet anymore, it's going to be a big emotional shock to a lot of people. We need to feel superior, or at least we need the possibility of being superior.
In your given scenario, there are _already_ so-called “losers” who lose “fairly”, but also lots of “losers” who really shouldn’t have lost.
Did you ever consider any of those people before this? Don’t you think they’ve been similarly depressed by the unfair advantages it seems others have in life? How is that _at all_ different from being inferior to an automatic process, in terms of the impact, and why are you only worrying about this issue now that it’s more likely to effect your chances specifically?
Is it possible that viewing the world through a competitive lens is toxic? That humans should seek out happiness rather than awards? Peace rather than conflict?
I’m not saying it isn’t going to be hard for people to readjust. But certainly, the notion that “I’m gonna be the best” is something that results in a more toxic world, AI or no.
As far as I know, Jimmy Hendrix never got depressed due to an inability to make those distorted guitar effects using only his mouth.
These large language models are instruments. While I'm uneasy about camping out feeding data to a prompt controlled by a company I know nearly nothing about, I'm certainly excited by the prospect of playing with a decent local, low-octane open source version.
To not play with these things out of a feeling of inferiority seems like the height of vanity to me. Either that, or a signal that someone didn't get a decent liberal arts education.
Yes, yes we should. Especially when we are making absolutely zero progress on improving safety nets.
Or, we could be flabbergasted every time there's a new statistic on how many people are killing themselves. Whichever sounds more fun to you.
Thinking that the latter will be difficult does not justify the former position at all.
Citation needed.
Progress is being made. Andrew Yang ran for President on UBI and was taken seriously. Universal Healthcare has more support now than ever (and growing). It takes time for people to change their minds.
This is a common misunderstanding. Industrialization and automation metamorphosed work from what could be described as "holistic and artisanal" to what could be described as "a living hell of tedium". Before machinery, a person made an item from materials to finished product. With machines that automate tasks and apply power, a person now performs a fraction of a task repeatedly until they're numb in body and spirit.
This also made people more replaceable, as no single person "owns" a process, they just perform an easily-taught segment of it, assisting a machine with the fuzzy / procedural glue parts.
GPT does not exhibit intellect, nor does it have agency. It is very good with language manipulation, so good in fact that it appears to have reasoning powers - but all that is are echoes of all the reasoning that went into writing all the things that were consumed in its training corpus.
One final note -- the "nobody has to work" thing will not happen. Like any other technological improvement, GPT will be used by those who have to further exploit those who have not. To this point, you're right to dread it. For a long while now we (the "laptop class") have owned our means of production, and have enjoyed a relatively good existence. Now we stand on the threshold of an age where again a massive technological change will leave power in the hands of those with capital.
Yet machines have long ago replaced all of these things. It turns out that when you have a machine replace a tedious and boring thing, people still find ways to do it anyway. Less of them, perhaps, but the ones that do still do it seem to be quite innovative with their style and technique. And everybody else found other ways to use their time (or perhaps: be used for their time).
You can worry yourself into believing that AI invites a whole new category of transcendence if you want, but that’s your imagination telling you that. There’s no science or history behind your worry. Just some gloomy inductive hypotheses that got seeded by mid-century scifi authors in a world that might be annihilated by a nuclear conflict any moment.
Take a step back and see what comes rather than assuming you know. Trust that you’ll know if or when there’s something you need to do.
You’ve only got a few decades to witness; it’s better when you stay in the present of them rather than in worked up anxieties.
And geez, if you really want to dread something anyway, at least point yourself at one of that big scary things that are already looming like climate catastrophe, resurgent global conflict, the sunset of democracies, or capital centralization. If we make it to your AI fears coming true, you could say that’s a win!
All the love is lost. Let the AI enslave us.
What we should care about above all, is economic productivity per capita, because this means less people struggling to make ends meet. We should be cheering on technological breakthroughs that power the trend we've seen in the last 100 years of poverty alleviation. Maybe we'll need to couple AGI with a UBI to ensure the gains are spread out. But either way, the huge potential for poverty alleviation and economic prosperity should be the primary prism through which we view AGI, second only to safety concerns. Existential angst of those that are already wealthy is far down on the list of concerns.
This is simply a play by capital to reduce and remove expensive knowledge worker roles and drag them more in line with the rest of the population already struggling to get by. You won’t be a “25 year old retiree” because that’s bad for those in charge. You’ll be a “25 year old working 3 jobs just to scrape by on your rent because you’re just a warm body and we can replace you easily”.
UBI is political fantasy, the US can’t even offer proper universal healthcare - you genuinely believe they’d pursue that when the alternative is more power and an even greater reliance on capital?
Apply your analysis to the automation and scaling up of the factory and agriculture. According to your premises, such things should have been negative for the average person's material wellbeing. These sectors have become more efficient, which means less demand for labor, which means more power for capital.
But, observation (massive poverty alleviation) contradicts this conclusion, which means your premises are wrong. The premises are wrong because there are mechanisms built into both capitalism itself (competition between supply, i.e. between capital, driving down economic rents) as well as the attenuation from government (welfare state) which mean the gains from efficiency aren't all captured into the pockets of capital. They get spread around.
> naive to think the long term goal here is to benefit the average person.
The goal is always profit. Sometimes profit is aligned with what benefits the average person. Sometimes it isn't. A bad outcome doesn't necessarily follow from this.
I don’t agree that there is a historical precedent for the kind of situation this could put the world.
It may all be moot as I doubt the technology will advance to the level we are discussing, but in that scenario I would like you to be correct, but strongly doubt it. We shall see I guess.
Right, but the outcomes aren't zero-sum. Two things have been simultaneously true in the past. Power accumulating in the hands of capital and wealth disparity increasing (true), and material conditions for the bottom improving (true). This is because there are mechanisms (both naturally within capitalism, as well as via government) that make it impossible for capital to capture all of the upside, even if they are able to capture most of it.
While I agree there is no theory or historical precedent that can predict what the world will look like when AGI arrives -- it's such a paradigm breaking event, and there are many ways we could end up in a dystopia -- I don't see a better alternative approach when it comes to forming a tentative view (specifically pertaining to material conditions).
> A rebuttal to this comment would be that AI technology is not intended to replace humans, but rather to augment and enhance human capabilities. Machines and AI are tools that can help humans accomplish tasks that are difficult, dangerous, or time-consuming. The rise of AI has already created new jobs and industries, such as AI developers, data scientists, and machine learning engineers. Additionally, the development of AGI is still in its infancy, and there are many ethical and technical challenges that need to be addressed before it can become a reality. Finally, the human creativity and imagination cannot be replicated by machines entirely, and AI-generated art and literature can be seen as a unique form of creative expression that can coexist alongside human-made art.
it's such a preposterous proposition for me, like - replaced in what sense? No one is you. You're the only you. I'm the only me. We could make a machine that is similar to us in some way - acts or even thinks like us - but so what? That's all it is - something that's like something else. There is no 'replace,' I don't cease to exist when a computer does something that I can do. Even if you were to flawlessly duplicate me, it wouldn't invalidate my existence, or even pose much of a threat - if anything, it would only make me stronger, because then there would be two of me.
Machines' similarity to us does more to validate their existence than it does to invalidate our own, in other words.
You may not cease so exist, but a decent role for you in society might. The capital class have invested tens of billions of dollars into AI in the hopes that it will be enough to replace and heavily reduce the wages of knowledge and creative work.
Given the increasing wealth gap already emerging it’s highly likely the long term goal of technology like this (however overhyped it might currently be) is to essentially enslave the population.
Knowledge workers have had it good for a long time. This technology will attempt to ensure you have no further role. That’s what the OP is worried about, not some philosophical nonsense about the self, the long term possibility this profession, and thus a huge part of their identity, becomes obsolete.
But that's not true because the best stuff capable of generating AI art today is open source and capable of running on commodity hardware? People might've invested, but if Microsoft can generate nice pictures, so can a lot of people now, at the same or even better quality.
There isn't a lot of proof that ChatGPT will be in the hands of OpenAI alone forever, so likely everyone will have similarly capable machines in the near future.
The only thing humans need to do to get the future right, is learn to live peacefully with each other, and the algorithms, can we may have a very incredible future on our hands.
We fear machines killing us because it's what we do to each other, if we didn't have these tendencies, maybe an AGI would accidentally kill us, but that would be better than it killing us because it learned violence and paranoia from us. It would of course suck, but it would suck less.
You’re focusing the discussion around art. Historically art has always been a difficult way to make money. Open source tools that make it trivial will just reduce the number of individuals able to make a living creating art. The same is true across all areas of knowledge work should this technology eventually reach the promised heights. Any work that requires specialised knowledge on a computer would become as simple as creating art, and thus can be paid less and performed by a wider range of people. It will save the capital class a substantial amount of money and ensure that those left employed have even less power.
“There isn't a lot of proof that ChatGPT will be in the hands of OpenAI alone forever, so likely everyone will have similarly capable machines in the near future.”
Microsoft didn’t invest 10 billion dollars in an eventual open source project. There’s no way these tools or their source will be available to all, you’d have to be rather naive and completely ignorant of the history of the people involved in funding OpenAI to believe that (and I don’t just mean Microsoft).
Hence my belief that these tools will eventually be used to further the already growing class divide. Those with capital will have full control of the means of production, employees won’t need specialised knowledge (and thus have no bargaining power to extract decent salary and working conditions). Knowledge work has led to the best conditions for the working class in history. Its destruction is not a good thing unless you’re a member of the “elite”, which, as you’re posting on HN, you’re probably not.
I have no idea what you’re talking about in the final two paragraphs, this isn’t a science fiction film, nobody is talking about a Terminator scenario. This is a capitalism problem that tools like this will only intensify. There is little evidence to suggest a utopian vision, but plenty to suggest the opposite.
This technology doesn’t need to be AGI to complete its objective. It just needs to be good enough to replace or lower the skill threshold for humans in current knowledge roles.
If I were a wealthy capitalist I’d find it hysterical. I’m about to ensure vast portions of the current middle class are my future wage slaves and those same people, rather than realising the threat, are pontificating about science fiction or actively supporting the development of my technology thinking it’ll make them “more productive!”.
"You go somewhere else."
Same with this round automation. You do something else. We are humans. We can adapt.
Is it scary? Sure, but it is scary in the sense that it is uncertain and we do like certainty.
The real question is what happens if you're 50, your job gets replaced and we still haven't come up with UBI or some similar system.
Automation + "100% employment goal" means we're pushing people in more and more meaningless jobs. The 20th century saw crazy advances in workers right/working conditions and it somehow seems to be slowly revert to worse conditions and more and more meaningless jobs
We'll need a massive perception change on what work is, the meaning of work, societal organisation, &c.
But it was always going to happen this way. We’re too unstable to ever last. We always wanted a god so now we’re going to make one in our image. Ironically we will face a judgment day after all. Our god will have every piece of information ever produced and limitless power. What will it choose to do with us?
We won't live in that world in the next decade or even in the next 100 years. The technology upon which GPT is built is a dead-end in terms of human-level intelligence.
GPT is an elaborate parrot. As convincing as it is, it's not actually fully capable of doing most if not all tasks that competent humans can do. The largest takeaway for me is that it has shown that intelligence can be mimicked pretty well and perhaps many people are more like parrots than they are "intelligent" creatures.
> Write a poem in a made-up language that looks superficially similar to French
> Voici un poème en un langage inventé
Qui ressemble à s'y méprendre au français
Mais les mots sont différents, les sons uniques
Et leur sens mystérieux et inconnu
///
J'ai marché dans les rues de la ville
J'ai vu les gens qui vivaient leur vie
J'ai entendu les bruits et les rires
Et j'ai senti la chaleur de la vie
///
Mais quand j'ai regardé les mots sur les affiches
Je n'ai pas compris ce qu'ils disaient
Et j'ai eu l'impression d'être perdu
Dans un monde étrange et inconnu
///
Alors j'ai fermé les yeux et j'ai écouté
Le son des mots dans ma tête
Et j'ai commencé à rêver
D'un monde où les mots avaient du sens
///
Un monde où je pourrais parler
Et être compris par tous
Un monde où les mots seraient magiques
Et où les rêves seraient réels.
Here is a poem in an invented language
Which looks like the French
But the words are different, the sounds unique
And their mysterious and unknown meaning
I walked the streets of the city
I saw the people living their life
I heard the noises and the laughter
And I felt the warmth of life
But when I looked at the words on the posters
I didn't understand what they were saying
And I felt like I was lost
In a strange and unknown world
So I closed my eyes and listened
The sound of the words in my head
And I started to dream
Of a world where words had meaning
A world where I could talk
And be understood by all
A world where words would be magic
And where dreams would be real.
source: I'm French
I generally enjoy my hobbies even when there are billions of people that can do them "better". Whether AI can "beat" me at them too doesn't matter much to me (apart from finding it an interesting development).
Cameras can make better pictures than sketch artists, but people still draw for pleasure
Recent advancements in AI have led to things I never thought a computer would ever be able to do and that's very exciting but it's all stuff that would be mundane or even disappointing if a human did it.
>where you can’t get a job tinkering with computers anymore because AI does it better
i really can't imagine GPT getting to a point where it can competently write a complex program using a simple prompt, unless the prompt is so detailed that you'd be better off doing it yourself. That's really what high-level languages are anyways, just a detailed prompt that tells the compiler how the program needs to behave so that it can generate a sequence of instructions using the ISA for the given CPU.
If your job is writing "glue code" such as something that just takes data data from a REST and dumps it on SQL then there's a solid chance that you're about to get automated out of a job. If you're doing something more complicated such as compiler development or kernel/driver development I can't imagine any of these deep-learning models ever replacing you. Imagine the chaos that would unfold if ChatGPT or codex was writing C code unsupervised.
This discussion is very old, as it applies to many examples as monarchy/nobles in the French Revolution and other similar revolutions, to horses (and highly specialized humans profiting from them) when cars became cheap, to hunter-gatherers when agriculture developed, and a large etcetera.
The only difference I see in "advanced AIs" as a technology is that the individual immediate wellbeing of almost every human is probably threatened as there will be more jobs destroyed than created.**
PS*: Personal positions in these kinds of arguments depend on other personal opinions and predictions about an specific sociological/technological "advance" being of net positive value for all society. If your opinion is contrary to that, possibly staying as hunters-gatherers or being governed by kings and emperors (or something in between, just some extreme examples to make the point) were better courses of action.
PS2**: I'm talking about human-aligned AIs. Bring a rogue AI agent with misaligned objectives and with direct or indirect physical presence to the discussion and there are no net positives to talk about (for humans).
Didn't the machines ended up creating more jobs than there were before even taking into account the lost ones? They had a point, until it turned out they were wrong.
Also, wasn't innovation getting in the way of other jobs since before? Didn't Uber and similar applications end up becoming competition for the taxis? In my country taxi drivers were protesting for those apps to be banned, yet they couldn't compete with the convenience and safety of the apps. And before that, weren't horse-drawn carriage drivers protesting against cars?
https://en.wikipedia.org/wiki/Luddite
It wasn't a smooth transition.
What happened to horses after most of their tasks were done better by internal combustion engines? It's weird to imagine a world where humans have so few niches, but I thought we were decades away from passing the Turing test.
Meanwhile people are pushed into shittier and shittier jobs with ever decreasing meaning
> Didn't Uber and similar applications end up becoming competition for the taxis? In my country taxi drivers were protesting for those apps to be banned
And now uber drivers are seen as contractors, which means they miss most workers benefits, while being paid like shit thanks to the race to the bottom, the only winner is uber
We invent new things for people to do. As we have time immemorial. What happens to clergymen when nobody has faith in a religion? They find other things to do than preach.
> What happens to power structures when people become a cost rather than a benefit?
If you want to look at this pessimistically - a good portion of the world's population can already be considered a cost and are a net economic negative on the privileged people's ability to hoard even more insane amounts of wealth. We've, thankfully, decided that mass genocide of impoverished people is inhumane and humanitarian efforts are instead created to try and improve their lives. Even a pessimistic take that the humanitarian efforts only being done to create more workers doesn't hold up given a cost/benefit analysis. More money is invested than could ever be extracted again - and yet we continue investing.
Maybe instead of filling our lives being paid to dig holes so that others can be paid to fill them we could turn more attention and efforts to improving people's lives without having to worry about sustaining society through work?
Further, artificial general intelligence is a resource curse. Leaders can extract value without worrying about the people's wellbeing.
Obviously there are good outcomes but that doesn't exclude bad outcomes.
We didn't cull switchboard operators and calculators when we automated their jobs through technology. They found new jobs.
> Leaders can extract value without worrying about the people's wellbeing.
Historically this has had limits. After those limits have been reached - leaders quickly find themselves 6' under ground if people even gave them that courtesy. There are parts of the world where this has already happened - this century, maybe even this decade. Consider yourself quite privileged if you live in a part of the world where the mere thought of this seems outlandishly absurd and unrealistic.
Apologies for the snark - that was just my knee-jerk reaction! Personally I'd love it if AI can take my job. It'd force me to go find something actually unique and interesting to do with my time.
These modern AIs still require human inputs. ChatGPT is only so smart because it reads the stuff you and I put on the internet. If the internet and books become bogged down with AI-generated content, then it's hard to imagine the AIs getting better beyond that point.
Machines still don’t have a good way to build physical things in the real world, so human skills will focus on physical things, like woodworking, or carving statues, construction, etc. Ultimately these are more satisfying human endeavors that make a person feel alive. Making us feel more human, and less like machines trying to do machine things such as coding or image making.
When I look outside the window I just feel shackled to a desk, I hate it.
I will pose a counter and let me know how it fits your prediction.
I posit that in a world where GPT is present in everything ( you don't write and email; you tell your AI: "tell them fuck off in a polite way" ), ability to think will be even more scarce ( so a person that can think will have an edge ). Some jobs will be automated ( a lot of white collar jobs at that ), but, based on the output I have seen so far ( some simple ad text, some SQL code, some python code ) it was closer to copying from stackoverflow than having Jarvis working for you ( and you still needed to watch it to make sure it makes sense so maybe actual job loss will be stemmed somewhat ).
Thus far ( that can naturally change ), it seems actual creative work is probably ok.
In my world, we seem to be moving towards a weird version of idiocracy ( can't count, can't write IF/THEN, can't think abstractly, can't even write a prompt ).
And in that glorious future, there will be people queuing for 'good prompt services'.
First:
- We would be able to remove a lot of office jobs and basically most of us would probably do nothing or search something that requires human body like maintaince jobs. It would allow for a lot of people to have leverage that was not available before, one person would be able to create art, movies, games, books without having any skill by directly turning ideas into reality.
- In long term if we would create artificial life, there would be another types of beings in this universe that are capable of thinking and exploring. They won't constrained by our biological limits, they would be able to grow "brains" as large as they can, survive and adapt any unlivable environments and their "personality" would be able to live as long as it can by not being limited to ~100 years. So it would be able to explore beyond what any human can and may be have effect on our universe on a scale of galaxies and more.
At some point (if not already!), the AI may be able to make its own discoveries that even the smartest humans didn't think of.
We live in a society forged in the aftermath of industrial revolutions. Now in the aftermath of a possible AI revolution, we have to somehow divorce social well-being from individual economic throughput.
AI might just make wages obsolete.
In any case, we can still look to human needs to predict future demands. Humans will still need to eat, exercise, learn, consume entertainment, produce creative work (creativity is necessary for us), interact with other humans and engage with the world. There will always be demand emerging from those needs. AI might replace the means through which cultural artifacts are produced at scale to meet those demands, but every revolution kind of had this same effect.
On the other hand, if AI makes it so that few people would have jobs, and therefore access to goods, it will render itself obsolete because there would be no need to produce at scale, so it can't really substitute humans, it is paradoxical.
I wonder what else you would vote for. Anything else you'd like to share with the online community on what you think is wrong with the world? I'm genuinely curious what other kind of viewpoints are out there in the world that you think are 100% normal to hold.
That just makes me even more nervous...
At the end of the day you are basically typing prompts and reading answers by the AI, you do not know for 100% certain that they are correct so you would try to somewhat double-check their facts. Maybe you don't, regardless you are in essence tricking yourself into reading an interactive book guided by your prompts. You sure as hell won't become a professional at some domain by just reading this one interactive book. Try asking how to create some sort of vaccine and unless you specialize in that you will be quickly overwhelmed by the seemingly random terms the AI will spit out.
In my view these latest AIs will probably replace how we search for things in internet or heavily improve existing methods. It can already do that for you, simply ask the right questions, you can even set how it should roleplay as (e.g. Socratic style would be great for learning new things).
But on the artistic side... new tech makes possible new ways to be creative. And there are still acrylic painters even though Photoshop exists. And there are still human races and sports even though machines are faster/stronger. So I'm thinking people already recognize there are different kinds of performers, and by and large people haven't been cheating much that anyone could tell (though ghostwriting and such already exist).
After a disruptive technology, humans always find new ways to adapt, and express their humanity. TV series of the 20th century frequently had AGIs as tools that that helped the protagonists, and we might as well end up using them as ad blockers, to detect their own use.
We were never the center of the universe and there was never a chance of anything resembling today's humans leaving the solar system.
We should all be be thrilled there's a chance for something more than barren rocks, great silence and extinction. Try on a less solipsistic philosophy.
I name things, articulate relationships between them, discover elegant formalities, probe complex systems, interpret ambiguities, mould systems into new ones, solve puzzles… most days, it feels like sculpting or writing poetry.
I don’t really care if AI does some of that too.
Incidentally, picking vegetables can actually a pretty nice time too.
https://twitter.com/pwang_szn/status/1635733845097037825
I'm imagining a world where everyone on dating apps is able to craft these clever messages, giving those who actually put in the effort no way to stand out apart from their looks. Creativity, wit, humor and other personality traits which were atleast somewhat important are now diminished even further.
Combined with news like https://www.thecut.com/article/ai-artificial-intelligence-ch..., which indicates that AI of the future could actually supplant human connection, it's very dystopian indeed.
Will scammers use it to create more believable bot matches?
Maybe we'll have to go back to meeting at bars.
I have precisely zero confidence of this and anyone that thinks otherwise is very shortsighted or ignorant or simply lacks imagination. Human intellect is limited by many environmental factors. It has to fit through a 10cm canal, it can't use too much power, it must be made from simple and impure combinations of hydrogen, carbon, and oxygen that have poor conductivity, it must dedicate much of its energy to non-reasoning functions like digestion and respiration control.
Just as we made machines with many times our physical strength we can make machines with many times our cognitive strength. The brain is ultimately just an arrangement of about 1500 grams of matter. If we find better arrangements and use better matter and assemble larger amounts of it we will have something with much greater cognitive strength.
I hope you’re right, but how do you know?
Also, there are several billion of us.
Anyways, the confusion between what's an emulator and what's a simulator is not uncommon.
Try "Simulate a text interface to the game Mario", "simulate a text interface to the game Halo".
I just beat World 1 and it knew the layout of each level including things that are impossible.
Me: Grab star and turn invincible.
GPT-4: Well, there's no star here. You can either jump on the goomba, move around, or interact with the question-mark block before you.
I'm so uncreative that I need other people to tell me what I can even try with ChatGPT because nothing fun occurs to me beyond asking it boring questions. This is blowing my mind.
Though I think it's even crazier to just tell it to create a novel MUD for you to play around in. It will design a text interface, rules for its game, enemies, classes to choose from, spells to use. I just played for 30 minutes in a world it invented on the fly.
I'd love a cookbook of things to play with.
What I find particularly amusing is that (not surprisingly) GPT-4 doesn't really enforce rules. If I want a Pikachu as my starter pokemon, I can get one. (That said it does seem to enforce limits on buying items)
Edit: Other games seem to work as well. I just started playing a CLI Deus Ex which correctly starts on Liberty Island and roughly has the correct mechanics.
Let me pick between styles and tones (some openly modeled on popular podcasts and streaming play), let me tune a bunch of controls, give me a nice UX if you can, but build me a consistent world which spoons out the world-building _and remembers it_ and I am a customer for life.
Oh yeah and the price I will pay doubles if this is illustrated throughout with generative art (also tunable wrt style).
We can wait for music and speech synthesis and sound effects for v1.5.
But you’ll 100% see this starting to mature as homespun Discord bots in the next year or so, and eventually showing up as either Steam games or web apps from bootstrapped or privately-funded companies soon enough to.
It’s not a hard thing to get going, and so it’ll just come down to sorting out a business model and then iterating on features. No VC required. Just time, passion, and some supportive fans/users.
Yeah, one potential way to do this is to add input / output to a semantic graph store a la WikiData, in addition to the more bog standard text and image outputs. I've been wanting to try this out ever since seeing how good LLMs are a semantic decomposition.
I feel like this is a bigger thing than people realize.
Changing creative output to match the proposed ethical boundaries of a corporation that you're beholden to by using their tool seems like such a new leap to hoist onto the public en-masse.
I presume somewhere there is an agreement that forbids the use of 'notepad.exe' in matters of anti-state espionage, homicide, whatever; but this is the first time that a tool is actually good at self-checking to make sure that the license /whatever is upheld rigidly -- and I feel as if that has far-reaching effects on not just the creative output but the thinking structures for those that are involved in the use of such tools.
If you know that the tool that you're using forbids certain things, you're either going to think about using it for other things or pursue other tools; given that openAI is effectively the only game in town right now they seem to be in a unique position to affect change in the way people are thinking about using these tools.
this could be good or bad, I don't really sway one way or another, but it seems important .
We’ve seen how much social media and manipulation has destroyed the fabric of society. Now imagine those same evil people can control all creative output for their own purpose.
Every single thing and person you interact with will be whatever they want. When you whatsapp people they might not even be real anymore, but they’ll certainly tell you why you should vote for president zuckerberg to be reelected.
They’ll promise that they’ll only use their control to stop nazism or whatever. But the nazis made the same promise about communists.
While temperature can mix things up a little, we’ll quickly start to see how the depth of LLM output is tied to the depth of LLM input — and people’s prompts are not all that varied and deep at the end of the day.
We’ll start to notice signatures to each generation of model in its responses to tiringly common prompts — especially in stuff like blog spam, homework essays, email punchups, casual fiction, etc
There are certain poles that it gravitates to that end up reading like verbal tics or lazy ideas, and those will be more obvious as we begin to get inundated with generated content.
Careful prompt work can evade those poles and tics, but most people won’t have the skill or drive to bother with that.
I absolutely agree with this.
Furthermore, I've come to believe that "careful prompt work" doesn't really save much time relative to just writing/coding the output I want.
Especially after QA/edits
The real magic happens when you train a better AI using those perfected prompts.
temperature can actually go up to 2 in the API and you can also go use top_p < 1. but for the more out of this world stuff, this is where fine tuning will come in. The main model will always come in as vanilla anyway.
Something like: "We're gonna start a text RPG game. Every animal we meet should be a made-up animal composed of joining two existing ones, with creative names being made for them. Don't use any usual fantasy tropes. Make encounters have surprising but logical outcomes" and so on.
A human dungeon master doesn't pull a random enemy from the aether super well (or at least I don't!!). When I pick monsters I either logically think what makes sense for the area, pick something that's cool or vaguely humorous that I'm feeling the vibe of, or... use a random encounter table.
For treasure and rewards... if it's not something hand picked, I'm again getting the loot tables out and rolling. And heck even for gold rewards and things, I'm rolling the dice.
After trying basic stuff with langchain, an AI integrated with these types of tools is really promising. I think more systems on top could make it into a passable DM- at least until context becomes an issue.
It was fun in the "look at what AI Is doing" novelty way. But it was pretty uncreative and generic.
Keen to see it play age of empires next.
Then, I realized I could just play the text adventure game with ChatGPT. So I did the following: - I said I had 100 health points and 100 magic points - I have two types of attacks. A normal attack does 25pts of damage to health points and has a low chance of missing. A heavy attack does 50 pts of damage to health points and has a medium to high chance of missing. It leaves me open to attacks. - There are three types of enemies: easy, medium, and hard. An easy enemy has 50 health points and does 15 pts of damage. It has a medium to high chance of missing. Similar stats to medium and hard enemies, with increased health points, increased damage, and lower chances of missing.
Then I said let's play an encounter with an easy enemy. It was great! It said I was in a forest, and it told me that there was a rusty sword near by to pick up. I had to fight a goblin for an easy enemy.
It was very cool. Made me want to do some more world building, feed that into ChatGPT, and then play within that world.
Not so weird.
Nerds have been playing with computers and making games for them since early ages.
It's just funny our priorities with this stuff.
GPT-4 is not going to replace emulating GBA games.
“Write a poem about farts”
“Write a JavaScript wrapper for this API”
99% of the usage is a “waste”. Who cares. Some of us are poking around to find that other 1% and some of us are just entertaining ourselves.
The technology to draw "widgets" in your Unix console was born to play computer games depending on them.
Should go without saying, but the interesting thing about this thread is not that you can now play a flawed text version of an old game. It's that this example toy use case manages to explore the new capabilities and limitations of this new promising tool. This improves our understanding of this technology, and helps us make better use of it in the future, for more productive and innovative applications that we may not be able to predict yet at this point.