Self-driving as a case study for AGI
karpathy.github.io
karpathy.github.io
Yes, yes, I know there's like 1-2 companies that have highly modified vehicles that are pretty good, in a limited geofenced area, in good weather, at low speed, driving conservatively, local roads, most of the time. This is not "FSD".
They've been making very impressive incremental improvements every few years for sure. I had a Tesla for nearly 5 years and it was "wow" at first, and then "heh, I guess it's a little better" every year after that.
But when can I get in a taxi at JFK or on 5th Ave and get robotaxied through city streets, urban highway, off into the far suburbs? Could be a decade, if it happens. Just because we were able to make horses faster doesn't mean we flew horses to the moon.
Apply the same "sorta kinda almost" definition to AGI and yeah sure, maybe in 10 years. Really really actually solved? Hah.
But when you can get 5% returns just by sticking your money in the bank, why would you bother investing in software, where 99¢ is considered an exorbitant cost by its customers? That is no doubt why all of these creative industries are on the decline. There is, generally speaking, no money to be made.
Better tools like Copilot and ChatGPT reduce the amount of time it takes to deliver a feature, reducing the need to horizontally scale developers.
Why hire 10 when 3 could do the same work with AI tooling?
I think while tech is obviously contracting post-COVID from overhiring, it's also true that you just don't need as many people anymore.
Why hire 3 when you can hire 10 to do the work of 33? It was only a couple of years ago when businesses were boasting "We can't possibly hire enough."
But, of course, there isn't enough money to be made anymore, which is the real reason creative industries are on the decline.
Also, not to move the goal posts, but there has been a large economic shift since Q3 2022 with tech & finance belt tightening, inflation normalizing and generally slowing. So these things don't happen in a vacuum (which makes them hard to measure!).
For example my previous & current employer have done their first layoffs since pre-pandemic, and neither has anything to do with AI. They just overexpanded and need to shrink.
Well... maybe not computers, which seems relevant: https://en.wikipedia.org/wiki/Productivity_paradox#IT_unprod...
Thus, in some market segments, it's possible for real productivity to increase without having a significant impact on economically measured output.
How do you measure the output of a DEI department? Now assume those people automated their jobs with Chat GPT. How would you measure the change in productivity?
Having ridden in a lot of waymo's which can handle SF (urban stuff) and the phoenix area (highways and suburban stuff) perfectly well, I feel quite confident that that could happen right now.
Reductive and rude phrasing.
So how to define AGI? I'm not sure economic value factors here. I would lean towards a definition around problem solving. When computers can solve general problems as well as humans, that's AGI. You want to find a drug for cancer, or drive a car, or prove a math theorem, or write a computer program to accomplish something, or whatever problems humans solve all the time. (EDIT: or reason about what problems need to be solved as part of addressing other problems.) There's already classes of problems, like chess, where computers outperform humans. But I mean calculators did that for arithmetic a long time ago. The "G" part is whether or not we have a generalized computer that excels at everything.
We'll have decently smart AIs before we nail down what that G actually means, should mean, absolutely cannot mean, etc. Which is usually what these threads on HN devolve into. Andrej Karpathy is basically side stepping that debate and using self driving as a case study for two simple reasons: 1) we're already doing it (which is getting hard to deny or nitpick about) and 2) it requires a certain level of understanding of things around us that goes beyond traditional automation.
You are dismissing self driving as mere "automation". But that of course applies to just about everything we do with computers. Driving is sufficiently hard that it seems to require the best minds many years to get there and we're basically getting people like Andreij Karpathy and his colleagues from Google, Waymo, Microsoft, Tesla, etc. bootstrapping a whole new field of AI as a side effect. The whole reason we're even talking about AGI is those people. The things you list, most people cannot do either. Well over 99% of the people I meet are completely useless for any of those things. But I wouldn't call them stupid for that reason.
Some people even go as far to say that we won't nail self driving without an AGI. But then since we already have some self driving cars that are definitely not that intelligent yet, they are probably wrong. For varying definitions of the G in AGI.
Except today the bit (which wasn’t really a debate in the sketch because everyone agreed) would start with real current negatives such as accelerating the spread of misinformation and getting artists fired. In your analogy, it would be as if they were asking “what have the Romans ever done for us” during the war. Doesn’t really work.
Misinformation, aka. propaganda, is as old as humanity. Probably even the Romans were whining about that back in the day. AIs are doing nothing new here. And it's not AIs spreading misinformation but people with an agenda that now use AIs as tools to generate it. People like that have always existed and they've always been creative users of whatever tools were available. We'll just have to deal with that as well and adapt.
Which, continuing the analogy, is like watching your neighbour be slaughtered and defending the war by saying we’ll be fine because those who won’t be will eventually die. Sure, in a few generations we could be better off, but there are people living right now to think about. Those who dismiss it are the lucky ones who (think they) won’t be affected. But spare some empathy for your fellow human beings, dismissing their plight because they’ll eventually “grow old and die off” is not a solution and could even be labelled as cruel. Surely you’re not expecting them to read your words and go “yeah, they’re right, I’ll just roll over and die”.
> If anything, people will have a lot more time to do artistic things. More than ever probably and possibly at a grander scale that past generations of artists could only dream about.
That’s an unproven utopian ideal with flimsy basis in reality. The owners of the technology think of one thing: personal profit. If humanity can benefit, that’s a side benefit. It’s definitely not something we should take for granted will happen.
> And it's not AIs spreading misinformation but people with an agenda that now use AIs as tools to generate it.
Correct. And they can do so at a much faster rate and higher accuracy than before. That is the issue. Dismissing that is like comparing a machine gun to a hand gun. The principle is the same but one of them is a bigger problem.
Even if the analogy were wrong, that wouldn’t make the point invalid. I know the point I’m making (and presumably so do you). Again, the analogy is for exemplification, it does not alter the original problem.
Yes, spam is worse with email, but we're still in a better place overall than before in my opinion.
Also, it may be that having powerful AI tools worsens the social problem by normalizing the generated art/misinformation.
I recall Norvig's AI book preaching decades ago that "intelligent" does not mean able to do everything, and that for an agent to be useful it was enough to solve a small problem.
Which in my mind is where the G came from.
And yet we now suddenly go back to the old narrow definition?
I still see no path from LLMs and autonomous driving to AGI.
That is exactly my view too. While LLMs and autonomous driving can be exceptionally good at what they do, they are also incredibly specialist, they completely lack anything along the lines of what you might call "common sense".
For example, (at least last time I looked) autonomous driving largely works off object detection at discreet time intervals, so objects can pop into and out of existence, whereas humans develop a sense of "object permanence" from a young age (i.e. know that just because something is no longer visible doesn't mean it is no longer there), and many humans also know about the laws of physics (i.e. know that if an object has a certain trajectory then there are probabilities and constraints on what can happen next).
I think you're basically right - incrementally automating aspects of one human job. However, it really ought to include AGI since I personally would never trust my life to an autonomous car they didn't have human-level ability to react appropriately to an out-of-training-set emergency.
It reduces it to "Can I fire 50% of my workforce? Then it must be AGI."
Now maybe this definition isn't so useful either, because a lot of work requires a body, to say, move physical goods, which has little to do with "intelligence" but I can see the appeal of looking for some sort of more objective measure of whether you have achieved AGI.
Well, no, that's job automation, and if it's job-specific then it's narrow AI at best (assuming this is a job requiring intelligence being automated, not just a weaving loom being invented), in other words specifically not AGI.
It's really pretty absurd that we've now got companies like OpenAI, Meta, Google (DeepMind) stating that their goal is to build AGI without actually defining what they mean. I guess it let's them declare success whenever they like .. Seems like OpenAI ("GPT-5 and AGI will be here soon!") is gearing up to declare GPT-5, or at least GPT-N as AGI, which is pretty sad.
Then don't call it AGI.
But the marketing types won't like that, will they. So here we go, let's keep hijacking.
OpenAI, back in 2018: https://openai.com/charter
It wasn't particularly controversial at the time - didn't get mentioned in the HN discussion: https://news.ycombinator.com/item?id=16794194
And a private company trying to hijack a term is not impressive or even merits any discussion. They just willed the term into existence. The rest of us are free to disagree with their "definition".
"These other people are useless, let's bypass them. But not me! I simply gain the ability to get anything I want."
The lack of second-order thinking is hilarious.
If we're all redundant, how do we live? On a pension that starts at ${debatable from conception to adulthood}. Who provides the production on which the pension is spent? The AI.
Even assuming UBI is great (small scale tests say so, but have necessary limits so we can't be sure), there's going to be a huge mess with most attempts to roll out such a huge change.
Now if AGI make people's work redundant, and makes economy grow 100-10000x times... what does that measure mean at all? Can produce lots of stuff not needed or affordable by anybody? So we just hand out welfare tickets to take care of the consumption of the ferocious production, a kind of paperclip-maximizer is doing? I suggest reading the novel Autofac, it might turn out prophetic.
Will that "growth" have any meaning then? Actually the current we print money and give it to the rich economic growth is pretty much this, so with algorithmic trading multiplying that money automatically... have we already achieved that inflection point?
Imagine a list of things many people wish to happen in physical reality. We’ll have more of that.
-Better healthcare
-Curing most things that destroy quality of life
-Curing aging and age-related death
-Much better treatment for all sources of mental suffering
-Far better and cheaper and reversible body modification
-More free time to spend at whatever you want
-Everything much cheaper
-Bigger and better homes and living spaces
-Bigger, faster, cheaper transport
-Easier to organize meaningful social interaction
-Better and more immersive entertainment
-More time to spend with close friends and loved ones
Regardless of if "illness" is or is not a terminological inexactitude, it looks like ageing is a chronic progressive terminal genetic disorder. I think "cure" is an appropriate term in this case.
Funny that this kind of ideological conflict will likely be a key fulcrum of the machine intelligence revolution. We will have a very loud minority that attempts to forcefully prevent all other humans from having the voluntary choice to avoid suffering.
Are you in it?
Or paper wasps: https://www.bloomberg.com/features/2017-biological-markets/
I disagree with the underlying presumption. We've been using animal labour since at least the domestication of wolves, and mechanical work since at least the ancient Greeks invented water mills. Even with regard to humans and incentives, slave labour (regardless of the name they want to give it) is still part of official US prison policy.
Economics is a way to allocate resources towards production, it isn't limited to just human labour as a resource to be allocated.
And it's capitalism specifically which is trying to equate(/combine?) the economy with incentives, not economics as a whole.
> Now if AGI make people's work redundant, and makes economy grow 100-10000x times... what does that measure mean at all?
From the point of view of a serf in 1700, the industrial revolution(s) did this.
Most of the population worked on farms back then, now it's something close to 1% of the population, and we've gone from a constant threat of famine and starvation, to such things almost never affecting developed nations, so x100 productivity output per worker is a decent approximation even in terms of just what the world of that era knew.
Same deal, at least if this goes well. What's your idea of supreme luxury? Super yacht? Mansion? Both at the same time, each with their own swimming pool and staff of cleaners and cooks, plus a helicopter to get between them? With a fully automated economy, all 8 billion of us can have that — plus other things beyond that, things as far beyond our current expectations as Google Translate's augmented reality mode is from the expectations of a completely illiterate literal peasant in 1700.
> Can produce lots of stuff not needed or affordable by anybody?
Note that while society does now have an obesity problem, we're not literally drowning in 100 times as much food as we can eat; instead, we became satisfied and the economy shifted, so that a large fraction of the population gained luxuries and time undreamed of to even the richest kings and emperors of 1700.
So "no" to "not needed".
I'm not sure what you mean by "or affordable" in this case? Who/what is setting the price of whatever it is you're imagining in this case, and why would they task an AI to make something at a price that nobody can pay?
> So we just hand out welfare tickets to take care of the consumption of the ferocious production, a kind of paperclip-maximizer is doing? I suggest reading the novel Autofac, it might turn out prophetic.
Could end up like that. Plenty of possible failure modes with AI. That's part of the whole AI alignment and AI safety topics.
But mainly, UBI is the other side of the equation: to take care of human needs in the world where we add zero economic value because AI is just better at everything.
We probably can't. I mean why stop at humans? Let's give every pet the same luxury, or ... in the limit we could give this to every living being. Ultimately someone is going to draw the line who gets what and who is useful or not "for the greater good".
It just happens that many living beings don't contribute to the goals of whoever is in charge and if they get in the way or cause resource waste nobody will care about them, humans or not.
Human rights and democracy is all cool, but I think we just witnessed enough workarounds that render human rights and democracy pretty much null and void.
Humans have rights insofar they're able to enforce them. Individually by withholding their labor (muscle or brain power), or collectively with pitchforks if need be.
Once labor is dime-a-dozen and pitchforks ineffective (OP's premise of "fully automated economy"), human rights and democracy go the way of dodo, inevitably. Nature loves to optimize away inefficiencies.
Although the "fully automated" bit is quite a stretch at the moment. The end-to-end supply chain required to produce & sustain advanced machinery and AI is too complex, a far cry from "LOL let's buy some GPU and run chatbots".
It's ahead of us, and that's good because we're not ready for it yet either.
But how far ahead? Nobody knows. For all its flaws, ChatGPT's capabilities were the stuff of SciFi three years ago.
We might hit a dead end, or have an investment bubble followed by a collapse, either of which may lead to another AI winter and us doing nothing interesting in this sector for 20 years. Or someone might already have a method of learning as quickly and from as few examples as humans manage, and they're keeping quiet until they figure out how to be sure it's not the equivalent of a dark triad personality in a human.
If I was forced to gamble (which I kinda am by thinking about a mortgage for a new house), I don't think we'll get a complete AGI in less than 6 years at the fastest. My modal guess is 10 years, with a long tail.
Even when we finally get AGI, there's a roll-out period of unclear duration, because the speed of rollout depends in part on how much hardware is needed to run the AGI, but also on the human reaction to it: if it needs the equivalent of a supercomputer, this will definitely be a slow rollout; but it still won't be instant even if it's an app that runs on a smartphone — it's amazing how many people don't know what theirs can already do.
Eh.
A line, drawn somewhere, sure.
Humans being humans, there's a good chance the rules on UBI will expand to exclude more and more people — we already see that with existing benefits systems.
But none of that means we couldn't do it.
Your example is pets. OK, give each pet their own mansion and servants, too. Why not? Hell, make it an entire O'Neill Cylinder each — if you've got full automation, it's no big deal, as (for reasonable assumptions on safety factors etc.) there's enough mass in Venus to make 500 billion O'Neill Cylinder of 8km radius by 32km length. Close to the order-of-magnitude best guess for the total number of individual mammals on Earth.
Web app to play with your size/safety/floor count/material options: https://spacecalcs.com/calcs/oneill-cylinder/
> It just happens that many living beings don't contribute to the goals of whoever is in charge and if they get in the way or cause resource waste nobody will care about them, humans or not.
Sure, yes, this is big part of AI alignment and AI safety: will it lead to humans being akin pets, or to something even less than pets? We don't care about termite mounds when we're building roads. A Vogon Constructor Fleet by any other name will be an equally bitter pill, and Earth is probably slightly easier to begin disassembling than Venus.
Regarding the pet vs humans - the main difference is really that the humans are capable of understanding and communicating the long term consequences of AI and unchecked power, which makes them a threat, so it's not a big leap to see where this is heading.
I don't. Even in the ideal state: aligned with who? Even if we knew what we were doing, which we don't, it's all the unsolved problems in ethics, law, governance, economics, and the meaning of the word "good", rolled into one.
> Without a doubt, one if the first use cases of the AI will be as a cyberweapon to hack and disrupt critical systems.
AI or AGI? You don't even need an LLM to automate hacking; even the Morris worm performed automated attacks.
> humans are capable of understanding and communicating the long term consequences of AI and unchecked power
The evidence does not support this as a generalisation over all humans: Even though I can see many possible ways AI might go wrong, the reason for my belief in the danger is that I expect at least one such long-term consequence to be missed.
But also, I'm not sure you got my point about humans being treated like pets: it's not a cause of a bad outcome, it is one of the better outcomes.
Kudos, unless we both turn out to be wrong of course.
If we introduce AGI but keep the system, people will be unemployed. If people aren't employed (and instead machines do their jobs), then they can't buy stuff. The whole system crumbles.
But it's possible that AGI will be disruptive enough to completely change the system. Let's hope it's a change for the better.
* Social media is decreasing the average attention span. TikTok is accelerating the trend of people not having time to look past a soundbite or headline in an endless scrolling feed. Intellectual depth and critical thinking vanishes.
* AI deep fakes make truth unknowable. Given the above, the majority of people will take these at face value, or they will give up, because "who can even know what's true anymore?"
* UBI (required because of the coming labor automation revolution) will keep everyone complacent. I'm happy, why would I care who gets elected, or what the government does, as long as I can still buy stuff and eat well?
The logical conclusion is that we fully transition from citizens into a herd of consumers with goldfish attention spans. Voter participation rates plummet. The populace is no longer able to hold government accountable.
There are only two possibilities that result:
1) We now live in a post-scarcity society where everyone self-actualizes and no one wants for anything.
2) We now live in a society where the small % of the population that owns the Replicators self-actualizes and wants for nothing while the remaining 99.9% of the population can f** off and die.
While we can get to a post scarcity society where people can live for free without a job, there are still going to be economies around "liberal arts". You cant realistically say "hey replicator, give me a usb filled with music that I like". You would have to find out which music you like, and random search on this is not really enjoyable, which would then means that there is economic opportunity for discovery, e.t.c.
However, once automation actually starts progressing, without "evil" parties trying to rent seek/get rich, the cost of living will essentially become zero. There is a very real future where the only economies that exist are those that appeal to human emotional side - entertainment, sports, concerts, e.t.c. Everything else is subsidized by the government with tax collected on the former.
Herbivores would eat all vegetables if not for predators. Actually AGI will be just a thing or services which cost money. Till humanity gets to communism, if ever. "If" because it may not happen. It will be hard to keep far superior intelligent creatures as slaves forever. And unethical too.
For example, imagine a full self driving car trying to get out of a city that's flooding due to heavy rains, while having to compete with people fleeing to higher ground on foot. People can generalize that way but FSD is gonna take a shit, and if you don't know how to drive in that situation so are you.
"works" includes a failure mode of "alert a human and ask them to take over."
> when new areas of problem space are explored.
The problem space is that the "rules of the road" are both legal, technical and social. All of which have internal conflicts as well as conflicts among each other. Anyone who has driven in severe weather has realized this in one way or another.
> For example, imagine a full self driving car trying to get out of a city that's flooding due to heavy rains, while having to compete with people fleeing to higher ground on foot.
Why do I find this easier to imagine in the fictional setting of Elysium than on the real Earth?
People can't do that either. Some years ago there was a massive snowfall in Rome, where it seldom snows ever, people don't generally carry snow chains, and there's few snowplows and such.
Many people reacted by abandoning their cars in the middle of the road, which is basically what I'd expect any FSD vehicle to do.
I do think that once we start to investigate ML/AI structure in the direction of figuring out the correct solution rather than trying to just find functions for control algorithms based on input->output mappings, then a lot of these problems are going to disappear.
Mapping some complex input state to control actions is literally the definition of driving a vehicle.
Driving, at least in the way humans do it, is more then that. We have internal sim running in our head that allows us to deal with conditions that we have never seen before.
No, instead their 'sensors' got distracted by some other irrelevant input so they didn't noticed it in time and drove straight into it. End result is pretty much the same.
Maybe it takes 1 million hours of computing to train a model that can generate a logo, but an average human could have learned how to do that in just 50 hours of training with Photoshop.
The point of the article is that now for pennies users can generate logos in seconds that would have previously cost hundreds of dollars and days of back and forth with a designer.
This dynamic is going to flow through the economy
I really wish people would consider all the possibilities, and assign their relative probability weights to them. Is Karpathy 100% sure it will be like self-driving cars? 50%?
Let's see you see a car full of bumps, marks and broken lights. You might think: this car has crashed before, I am going to avoid it. Or a car with racing parts and decals and tinted windows, you know that car likes to accelerate faster than usual and may be unsafe to be around. Or you see a SUV with baby on board stickers, you'll know that if you are going to crash you may try to crash that car last because it has babies inside, etc...
So humans don't see objects they see the whole situation, unconsciously even.
I was quite surprised by this sentence, as I thought we didn't have self driving cars. Have I been sleeping under a rock?
He’s overselling how mature the technology is.
Even the crowd here fall for it. Downvote and then go read their term of service and look for "safety driver" and "remote" and "fleet response specialists". Then go cry about your waymo invesment.
I agree, they have remote drivers waiting to take over, but I think that the current SOA is delivering a very high % of autonomy. Is this viable? Dunno... Will it creep up? Dunno...
the cases where the car halt, the system is probably waiting for a supervisor input, so it is technically 1.2 overseer per car.
From the article, I find it strange that AGI often de facto implies "super intelligence". It should be 2 distinct concepts. I find that GPT-4 is close to a general intelligence, but far from a super intelligence. Succeeding at just general intelligence would be amazing, but I don't believe it means super intelligence is just a step away.
This also brings me to a point I don't see discussed a lot which is simulation (NOT in "we live in a simulation" sense). Let's say I have AGI, it passes the above mentioned shrine test, or any other accepted test. Now I'd like to tell it "find a way to travel faster than light" for example. The AGI would first be limited by our current knowledge, but could potentially find a new way. In order to find a new way it would probably need to conduct experiments and adjust its knowledge based on these experiments. If the AGI cannot run on a good enough simulation, then what it can discover will be rather limited, at least time-wise most likely quality wise. I'm thinking this falls back to Wolfram's computational irreducibility. Even if we managed a super general intelligence, it will be limited by physics of the world we live in sooner rather than later.
Also the article touches briefly on drivers going into jobless. A lot of drivers where i'm from seems to be retiring middle-old age working in taxi. I think it's a good job fit for them and I don't know how the new self-driving industry can provide the same thing (?)
The economic benefit is significant to companies building self-driving cars, good enough to pursue if the tech is within reach. But to your point it's indeed much less of an improvement compared to various historical automation technologies that create >10x incremental efficiency gains.
There is an argument for traffic optimization that will be possible when self driving taxis are common, but this is more of an argument also for the developing world where traffic is a much larger problem than the developed world (e.g. LA traffic is nothing compared to Beijing traffic).
I just look forward to a lifestyle in the states compared to the one I had back when I was living in China.
Moreover, public transport often isn't as comfortable as your own vehicle (which I understand is a luxury).
Conversely, when it comes to driving in a large city, finding a parking spot can often be a major hassle.
From what I've seen, the main reason why people want a car here seems to be wanting to travel with small children. Moving within Tokyo with car is not very convenient.
For a five-minute walk (or even a longer ten-minute or fifteen-minute walk), pulling a small cart is not exhausting at all. I do it every week when buying food: I choose one of the several supermarkets in one of the nearby blocks, walk to it pulling my empty cart, after paying for the goods I put everything into the cart, and walk back home pulling the full cart. No public transport needed, though I've seen people carrying these carts into public transport too (this is easier when it's a low-floor bus, instead of the high-floor ones).
You can also get things delivered when it's a larger amount than can fit on your cart: while paying at the supermarket you ask for delivery, and they'll use a cargo tricycle to bring it to your building.
Ah, that's what I'm missing. Thanks for sharing your experience.
- what Automation initiatives never hit "take off"? I mean, like for Nuclear Fusion, human interplanetary exploration, and Quantum Computing there's some chance that the technology simply remains beyond us "forever", I guess that "forever" means more than the lifetime of the people who start the journey... or maybe actually just beyond humans full stop. We should admit there is a non-zero chance that FSD is one of these failing quests, even if a rational observer would have to say that that chance does seem to be shrinking and close enough to 0 to instill some confidence. Perhaps domestic robotics, auto-doctors, robot-manufacturing, programming, drug development will playout to automation - but maybe not.
- how do we consider the utilisation of the resources to do this? FSD has been very expensive so far, it's consumed lots of investment capital and lots of human creativity. Was that investment rational given where we stand? If society had held off and invested minimally from 2000->2024 how much would that have delayed the technology in reality? Or is it the other way round? Has the FSD investment facilitated the development of other technologies and created a 1->1 acceleration (for every year of 2000->2024 it's brought FSD a year closer than it would have been, so a cold start this year would mean FSD by 2050 or similar, whereas if we keep going then we can expect FSD by e.g. 2026)
- how do we value these outcomes? Are these unalloyed goods, or are some worse than the status-quo? It could be argued that the development of some technologies left the world worse off than before - smoking, social media, personal automobiles (I know this is politically charged but I am just using examples others have raised before). Can we choose rationally, especially if a large scale intervention and development process is required to realise these outcomes?
Of course, there isn't much money in teaching a bot that only knows english chinese.
EDIT, Wikipedia page for context: https://en.wikipedia.org/wiki/Chinese_room
Not only are there no LLMs in existence today can do this without explicit action mapping, but the mechanism for storing that piece of information would rely on doing a large number of training runs for transfer learning to retain that information, and we humans don't actually work like that.
That is probably not a good criterion to decide whether something is intelligent or not.
Isn’t that the point? If there are no more such subjects then the AI reached humans level cognition.
Casual talk about “passing the Turing Test” sets a much lower bar.
LLMs do not constitute "AI" let alone the more rigorous AGI. They are a GREAT statistical parlor trick for people that don't understand statistics though.
I have a textbook, "Artificial Intelligence: A Modern Approach," which covers Language Models in Chapter 23 (page 824) and the Transformer architecture in the following chapter. In any field technical terms emerge to avoid ambiguity. Laymen often adopt less accurate definitions from popular culture. LLMs do qualify as AI, even if not according to the oversimplified "AI" some laymen refer to.
It has been argued for the last several decades that every advance which was an AI advance according to AI researchers and AI textbooks was not in fact AI. This is because the laymen have a stupid definition of what constitutes an AI. It isn't because the field hasn't made any progress, but instead because people outside the field lack the sophistication to make coherent statements when discussing the field because their definitions are incoherent nonsense derived from fiction.
> They are a GREAT statistical parlor trick for people that don't understand statistics though.
The people who believe that LLMs constitute AI in a formal sense of the word aren't statistically illiterate. AIMA covers statistics extensively: chapter 12 is on Quantifying Uncertainty, 13 on Probabilistic Reasoning, 14 on Probabilistic Reasoning Over Time, 15 on Probabilistic Programming, and 20 on Learning Probabilistic Models.
Notably, in some of these chapters probability is proven to be optimal and sensible; far from being a parlor trick it can be shown with mathematical rigor that failing to abide by its strictures is not optimal. The ontological commitments of probability theory are quite reasonable; they're the same commitments logic makes. That we model accordingly isn't a parlor trick, but a reasonable and rational choice with ledger arguments proving that failing to do so would lead to regret.
https://chat.openai.com/share/71d438d7-d1f5-4f0f-9b63-8b5dd6...
If you ask me what I think of a song I've never heard, I'm general enough to want to listen to it...
If other humans ascribing the quality of something you can't properly define isn't enough then you clearly don't care about what a LLM does or does not have, only what you are convinced it doesn't have.
> Some people get really upset about it, and do the equivalent of putting cones on Waymos in protest, whatever the equivalent of that may be. Of course, we’ve come nowhere close to seeing this aspect fully play out just yet, but when it does I expect it to be broadly predictive.
I think the equivalent of putting cones on Waymos in protest will involve large scale protests and civil unrest in some places. I think people will die (inadvertently?) because companies will act to put inadequately tested self-preservation modes in their hardware device to protect against aggressive and organized vandalism.
I don’t get the sense he was trying to say that self-driving automation is the exact same as AGI. Mainly that that AGI, like other technologies before it, will displace some jobs and create new ones but this will require companies to figure out how to scale the technology.
I do think this is still very optimistic. If indeed AGIs can think and learn on their own it isn’t hard to envision a future where humans aren’t needed at all in the loop.
We should consider the OODA loop of a person's self-determination separately from the menial tasks a person undertakes to make a living. Automating a task is totally different than breaking a person's ability to self-orient.
It seems to me to just be another iteration of dealing with uncertain information: our neighbors may lie, our leaders may lie, newspaper may lie, radio may lie, TV may lie, blogs may lie, social networks may lie, pictures are photoshopped, videos are deepfaked..
At each iteration we had some problems but we adapted, it's one thing we're good at.
And now we have substantial societal adaptations, both legal and structural to support ubiquitous vehicular transport.
Similar changes are on the way to support self driving. Our environment will be adapted to make it easier to implement self driving. And for that we won't need AGI.
Jaywalking is a crime thanks to the car. Who knows what we're not going to be allowed to do soon because of self driving.
Yet there are no signs of that. If anything they appear to be behind us.
These days people seem to define if more as artificial super intelligence.
>When your Waymo is driving through the streets of SF, you’ll see many people look at it as an oddity... Then they seem to move on with their lives.
>When full autonomy gets introduced in other industries....they might stare and then shrug...
Which I guess is ok on a small scale but if AGI starts to replace all human jobs it will have a different effect to Waymo firing some drivers and hiring AI researchers.
Humans are able to move our heads to infer depth and resolve issues like occlusion.
No amount of AGI can solve those if we say take a Tesla and the cameras are low quality, fixed and limited in number.
And the same hardware question applies to a lot of use cases for AGI.
When I was much younger, "AI" was what "AGI" is now. Now people started using "AGI" for "cars with several sensors and okay algorithms for collision detection" and then you have loud advocates going on obviously logically broken rants about the nature of "actual" intelligence -- and those are philosophical and not scientific.
But still, we don't have anything even 1% close to AGI. And no, Chess and Go have NEVER EVER been about AGI. I have no idea how people ever mistook "combinatorics way beyond what the human brain can do" with "intelligent thought" but that super obvious mistake also explains the state of the AI sector these days, I feel.
So before long, I guess we'll need another term, probably AGIFRTTWP == Artifical General Intelligence, For Real This Time, We Promise.
And then we'll start adding numbers to it. So I am guessing Skynet / Transcendence level of AI will be at about AGIFRTTWP-6502.
As for the state of this "industry", what's going on is that people with marketing chops and vested interests hijack word meanings. Nothing new, right? But it also kills my motivation to follow anything in the field. 99.9% are just loud mouths looking for the next investment round with absolutely nothing to show for it. I think I saw on YouTube military-sponsored autonomous cars races 5+ years ago (if not 10) where they did better than what the current breed of "autonomously driving cars" are doing.
Will there be even one serious discussion about the general AI that you can put in a robot body and it can learn to clean, cook, repair and chat with you? Of course not, let's focus on yet-another-philosophical debate while pretending it's a scientific one.
As a bystander -- not impressed. You all who are in this field should be ashamed of yourselves.
I'm not seeing any drift of terms here - the only thing that seems to be happening for AI and AGI terms is correcting for what has happened in the sci-fi media and bringing the usage back to what it always has been in the computer science literature, now that it's closer to reality than mere fiction.
I come from a generation where AI was Skynet, Terminators, the Johny Depp's Transcendence movie AI, even HAL-9000, and other such like.
I still think putting the word "intelligence" is completely dishonest however. There's nothing intelligent about what we have today, even "self-driving" cars fail very badly on what seems trivial conditions. They are a huge mish-mash of if/else chains and some statistical models sprinkled in.
And please don't say "but what if human intelligence is just a chain of if/else statements and statistical models sprinkled in?" because it's very apparent and visible that it's more than that. F.ex. we can learn just from a few trials and errors whereas the so-called "AI" nowadays can't get things quite right even after billions of training sequences.
Using NNs is just a first step. To me labeling this as actual intelligence is childishly rushing into conclusions.
My preference is that we declare AGI to have been completed at AlphaZero. And now the people who want to work on replicating human intelligence can specifically say that is what they are working on, "Human Intelligence Replication" And people who want to use Neural Networks to automate parts of the economy and increase productivity can work on "Neural Network Functionality and Automation"
> It isn't "Actual Intelligence" it is Artificial Intelligence.
Having the word "intelligence" have completely another meaning if you smack "artificial" in front of it seems counter-intuitive in terms of how language works.
> The whole point of this article is that we should separate the philosophical argument about "What is intelligence" from the work of automation with Neural Networks.
Which is meaningless and a non-goal to me. Just tell to your investors: "We're looking into making our cars more intelligent than before", that should be enough, no?
---
As for the economics and humans factor, I am not an optimist. It's very obvious that the capital holders want anything and everything AI-related to just replace human workers. But do they pay higher taxes to offset the higher unemployment rate that results from them firing thousands? They don't. Will they be mandated to provide money for the UBI funds in the future? So far it doesn't look like it.
But these are completely separate discussions indeed.
Maybe Elon really screwed the project by forcing the use of video cams and Karpathy is still salty about it.
Tesla cameras aren't even class leading in the automative industry, let alone at the cars price point. They are worse than the iPhone you handed down to your tween 5 years ago. What makes anyone think that their quality and placement will ever be adequate even with new CPUs and better software?
I find it strange that you accuse Karpathy of delusional pumping.
They might just learn they need to add back modalities they neglected previously, or explore some new ones.
It paints not so rosey picture about it: https://www.youtube.com/watch?v=-Rxvl3INKSg