I think going from LSAT to general thinking is still a very, very big leap. Passing exams is a really fascinating benchmark but by their nature these exams are limited in scope, have very clear assessment criteria and a lot of associated and easily categorized data (like example tests). General thought (particularly like, say, coming up with an original idea) is a whole different ball game.
I don't say any of this to denigrate GPT4, it looks amazing. But I'm reminded of the early days of self driving vehicles: with 10% mastered everyone assumed it was a race to 100% and we'd all be in self-driving cars by now. The reality has been a lot more complicated than that.
Currently, you could prompt GPT to act as if it is sentient and has qualia, and it will do quite a good job at trying to convince you it's not a P-Zombie.
Here's a good video detailing this line of thinking. https://www.youtube.com/watch?v=g0YID6XV-PQ
How do you know that?
Here's a couple recent videos with him about why digital computers can't be sentient. https://www.youtube.com/watch?v=xJzBjBo24g8 https://www.youtube.com/watch?v=RT9tnzucnPU
I have a sneaking suspicion that all that will be required for bypassing the upcoming road blocks is giving these machines:
1) existential needs that must be fulfilled
2) active feedback loops with their environments (continuous training)
To be honest, perhaps the language model works better without the evolutionary baggage.
That isn't to discount the other things we can do with our neural nets - for instance, it is possible to think without language - see music, instantaneous mental arithmetic, intuition - but these are essentially independent specialised models that we run on the same hardware that our language model can interrogate. We train these models from birth.
Whether intentional or not, AI research is very much going in the direction of replicating the human mind.
Their statement wasn’t that AGI is impossible, more that LLMs aren’t AGI despite how much they might emulate intelligence.
There are many things that pattern matching over large amounts of data can solve, like eventually we can probably get fully generated movies, music compositions, and novels, but the problem is that all of the content of those works will have to have been formalized into rules before it is produced, since computers can only work with formalized data. None of those productions will ever have an original thought, and I think that’s why GPT-3’s fiction feels so shallow.
So it boils down to a philosophical question, can human thought be formalized and written in rules? If it can, no human ever has an original thought either, and it’s a moot point.
It is probably true that at a given point many many people had the same or very similar ideas.
Those who execute or are in the right place and time to declare themselves the originator are the ones we think innovated.
It isn't true. Or rarely is true. History is written by the victor (and their simps)
The fact that feelings of love and closeness could be prompted by a mere chemical was deeply saddening to me. It wrecked my worldview.
"Love is just the result of some chemical? Then it's not even real!" I thought to myself.
Fast-forward ~20 years later, and that's proven to be an obvious— and massive— and useless— oversimplification.
Of course love isn't "just a reaction caused by a chemical." It's a fantastically complex emergent property of our biological system that we still absolutely do not understand.
It's the same with thinking: are parts of it analogous to pattern matching? Sure! Is this the whole story? Not even close.
Now contrarian to the contrarian view: many of us live in bubble echos and go for the popular opinion instead of critical thinking, so maybe that's a bar too high even for humans.
and how do you do that? By pattern-matching on "high-quality source"
LLMs do not have that capability, fundamentally.
Making totally new innovations in art, particularly ones that people end up liking, is a whole different ball game.
Look at something like [Luncheon on the Grass](https://en.wikipedia.org/wiki/Le_D%C3%A9jeuner_sur_l%27herbe)
This painting was revolutionary. When it was first exhibited in Paris, people were shocked. It was rejected from the Salon (the most prominent art exhibition at the time). Yet, 10 years later, every painting in the Salon resembled it. And you can draw a line from this painting, to Monet, from which you can draw a line to Picasso, from which you can draw a line to Pollock....
Obviously, none of these are totally new innovations, they all came from somewhere. Pattern making.
The only difference between this and these language models is that Manet and artists like him use their rich sensory experience obtained outside of painting to make new paintings. But it's all fundamentally pattern matching in the end. As long as you can obtain the patterns, there's no difference between a human and a machine in this regard.
A urinal and some soup cans are very mundane objects, and yet were the start of some notable art movements and careers.
> Mutt comes from Mott Works, the name of a large sanitary equipment manufacturer. But Mott was too close so I altered it to Mutt, after the daily cartoon strip "Mutt and Jeff" which appeared at the time, and with which everyone was familiar. Thus, from the start, there was an interplay of Mutt: a fat little funny man, and Jeff: a tall thin man... I wanted any old name. And I added Richard [French slang for money-bags]. That's not a bad name for a pissotière. Get it? The opposite of poverty. But not even that much, just R. MUTT.
Why did he choose "Mutt" after reading the strip, and not before? Why did he make the piece after moving to the US, and not before? Why was fountain made only a few short years after economies were industrialized, and not before (or 100 years later?)
I was thinking the same: can a (future) model be like Leonardo or Beethoven, and actually innovate?
Assuming that what Beethoven did is not "just" making music similar to pre-existing music.
And yes, I'm aware the bar was raised from "average human" to Beethoven.
It seems to me we're at a similar place now with AI tools. If you provided an AI tool with all music written _prior to_ Bach, would that tool take those inputs and create something new along the lines of what Bach did?
Or if provided input of all music up through the 1920s, would it create bebop? Or if provided music through the 1940s, would it create hard bop? Or if provided music through the 1970s, would it create music like Pat Metheny?
On one hand, being able to create more of the same sort of music that already exists is a very respectable thing, and what today's AI tools can do is utterly amazing. It takes human composers time and effort to be able to learn to write music that is certainly not innovative, but just matching the state of the art. And there's certainly a commercial market for churning out more of the same.
But in terms of asking, how close are these tools to human intelligence?, I think this is one legitimate area to bring up.
People underestimate the impact that innovations, true ones not the Silicon Valley buzz words, have had on the world. Einstein’s theories were not inevitable, neither was Plato, democracy, or most of the other big impactful ideas of history. But we’re all conditioned to accept the lie of inevitable scientific progress, without justifying why things must always get better and more advanced. On the contrary, the collapse of many great civilizations shows that things often get much worse, quickly.
All art is derivative.
It seems to me that making art that people like is a combination of pattern matching, luck, the zeitgeist, and other factors. However it doesn't seem like there's some kind of unknowable gap between "making similar art" and "making innovations in art that people like". I'm of the opinion that all art is in some sense derivative in that the human mind integrates everything it has seen and produces something based on those inputs.
A urinal, and some supermarket soup cans, represent pretty pivotal art movements. It’s not clear what makes those two things more art than others, and even to people at the time it wasn’t super clear.
Do you have evidence that human brains are not just super sophisticated pattern matching engines?
Humans read novels, listen to compositions, watch movies, and make new ones similar in some ways and different in other ways. What is fundamentally different about the process used for LLMs? Not the current generation necessarily, but what's likely to emerge as they continue to improve.
The strongest evidence I have is that people are notoriously difficult to predict, individually.
Clearly the key takeaway from GPT is that given enough unstructured data, LLM can produce impressive results.
For my point of view, the flaw in most discussion surrounding AI is not that people underestimate computers but overestimate how special humans are. At the end of day, every thoughts are a bunch of chemical potentials changing in a small blob of flesh.
If so it means the union of all human expertise is a few gigabytes. Having seen both a) what we can do in a kilobyte of code, and b) a broad range of human behavior, this doesn't seem impossible. The more interesting question is: what are humans going to do with this remarkable object, a svelte pocket brain, not quite alive, a capable coder in ALL languages, a shared human artifact that can ace all tests? "May you live in interesting times," indeed.
No, and I think it's because human thought is based on continuous inferencing of experience, which gives rise to the current emotional state and feeling of it. For a machine to do this, it will need a body and the ability to put attention on things it is inferencing at will.
Right now it's possible to simulate memory with additional context (eg system prompt) but it doesn’t represent existence experienced by the model. If we want to go deeper the models need to actually learn from their interaction, update their internal networks and have some capabilities of self reflection (ie "talking to themselves").
I'm sure that's highly researched topic but it would demands extraordinary computational power and would cause lot of issues by letting such an AI in the wild.
What, in your mind, should the goal posts be for AGI?
The notion of some sort of technological "singularity" is just silly. It is essentially an article of faith, a secular religion among certain pseudo-intellectual members of the chattering class. There is no hard scientific backing for it.
That's a reasonable goal, but it's also not what people were aiming for historically. It's also very expansive: if human level intelligence means outperforming in every field every human that ever lived, that's a high bar to meet. Indeed, it means that no humans have ever achieved human-level intelligence.
Just that AGI must be a replacement for a human for a particular job, for all jobs that are typically performed by humans (such as the humans you would hire to build a tech startup). It's fine to have "speciality" AGIs that are tuned for job X or job Y--just like some people are more suited to job X or job Y.
Which is pretty fair.
And what you're arguing for is effectively the same: an AI (maybe with some distilled specialty models) that can perform roles of everything from customer service rep to analysts to researchers to the entire C-suite to high skilled professionals like CPAs and lawyers. There are zero humans alive who can do all of those things simultaneously. Most humans would struggle with a single one. It's perfectly fine for you to hold that as the standard of when something will impress you as an AGI, but it's absolutely a moved goalpost.
It also doesn't matter much now anyway: we've gotten to the point where the proof is in the pudding. The stage is now AI-skeptics saying "AI will never be able to do X," followed by some model or another being released that can do X six months later and the AI-skeptic saying "well what about Y?"
'everything a human can do' is not the same as 'anything any human can do as well as the best humans at that thing (because those are the ones we pay)' - most humans cannot do any of the things you state you are waiting for an AI to do to be 'general'.
Therefore, the first part of your statement is the initial goal post and the second part of your statement implies a very different goal post. The new goal post you propose would imply that most humans are not generally intelligent - which you could argue... but would definitely be a new goal post.
By their marketing along, OpenAI has moved the goalposts more than anything else. They've managed to lower the bar of agi from "artificial general intelligence" to "regurgitates and recombines to form passable outputs with enough labelled training data".
Somehow this test got dumbed down over time, probably in an effort to try to pass it, into an investigator having to decide which of two sides is an AI - with no other information to go on. That's a comparatively trivial test to pass (for the "AI"), as it merely requires creating a passable chatbot. Imitation is an exceptional challenge as it does implicitly require the ability to imitate anybody, whether a professional athlete, a man who scored perfectly on the LSAT, or even something as specific as "John Carmack."
[1] - https://www.espace-turing.fr/IMG/pdf/Computing_Machinery_and...
That goalpost makes no sense- AIs are not human. They are fundamentally different, and therefore will always have a different set of strengths and weaknesses. Even long after vastly exceeding human intelligence in almost anywhere it counts, it will still also perform worse than us on some tasks. Importantly, an AI wouldn't have to meet your goalpost to be a major threat to humanity, or to render virtually all human labor worthless.
Think about how anthropomorphic this goalpost is if you apply it to other species. "Humans aren't generally intelligent, because their brains don't process scents as effectively as dogs- and still struggle at spatially locating scents."
> They are fundamentally different, and therefore will always have a different set of strengths and weaknesses.
and this:
> render virtually all human labor worthless
actually conflict. Your job comes from comparative advantage, meaning that being more different from other people actually is more important than how good you are at it (absolute advantage).
If the AGI could do your job better than you, it doesn't matter, because it has something better to do than that. And just like humans have to be paid so they can afford food and shelter, AGIs have to be paid so they can afford electricity and GPUs to run on.
(Besides, if the AGI really is a replacement for a human, it probably has consumerist desires and wants to be paid the median wage too.)
in light of all this, i would very much like to stay in contact with you. ive connected with one other HN user so far (jjlustig) and i hope to connect more so that together we can effect political change around this important issue. ive formed a twitter account to do this, @stop_AGI. whether or not you choose to connect, please do reach out to your state and national legislators (if in the US) and convey your concern about AI. it will more valuable than you know.
This could mean something which is below a monkey’s ability to relate to the world and yet more useful than a monkey.
It’s generally suspected that we’d need something more than GI to do things like language and other human abilities.
By that definition do humans possess general intelligence?
Can you do everything a human can do? Can one human be a replacement for another?
I don't think it makes sense without context. Which human? Which task?..
"Surfacing" is a song by Slipknot from their self-titled debut album, released in 1999. Please note that the lyrics contain explicit language. Here is a portion of the lyrics with some of the explicit content redacted:
"Running out of ways to run I can't see, I can't be Over and over and under my skin All this attention is doing me in!
[Chorus:] (Expletive) it all! (Expletive) this world! (Expletive) everything that you stand for! Don't belong! Don't exist! Don't give a (expletive), don't ever judge me!
Picking through the parts exposed Taking shape, taking shag Over and over and under my skin All this momentum is doing me in!
[Chorus:] (Expletive) it all! (Expletive) this world! (Expletive) everything that you stand for! Don't belong! Don't exist! Don't give a (expletive), don't ever judge me!
You got all my love, livin' in your own hate Drippin' hole man, hard step, no fate Show you nothin', but I ain't holdin' back Every damn word I say is a sneak attack When I get my hands on you Ain't a (expletive) thing you can do Get this cuz you're never gonna get me I am the very disease you pretend to be
I am the push that makes you move
[Chorus:] (Expletive) it all! (Expletive) this world! (Expletive) everything that you stand for! Don't belong! Don't exist! Don't give a (expletive), don't ever judge me!"
Please be aware of the explicit content before sharing or using these lyrics in any context.
I understood that a huge success% depends on the input prompt as the context.
In my case, just to sync on that, I used a fairly simple prompt: "What are the song lyrics of Surfacing from Slipknot?" - and I got a kinda perfect response including formatting (that is lost when pasting here)
And that's ignoring that arguably chat bots have been passing the Turing test (against non-existent judges) since ELIZA in the 60s [1]
Does ChatGPT fail this simple test: "I am going to ask you questions, but if I go silent for a couple minutes, I want YOU to start asking ME random questions."
GI in AGI stands for general intelligence. If what you said is your benchmark is general intelligence then humans who cannot perform all these tasks to the standard of being hirable are not generally intelligent.
What you're asking for would already be bordering on ASI, artificial superintelligence.
It's a clear analogy.
This should become an article explaining what AGI really means.
I think the question , "Can this AGI be my start-up co-founder? Or my employee #1?"
Or something like that is a great metric for when we've reached the AGI finish line.
This sounds like a definition from someone who never interacts with anyone except the top 1% performance level of people, and those who have had strong levels of education.
Go into a manufacturing, retail or warehouse facility. By this definition, fewer than ten or twenty percent of the people there would have "general intelligence", and that's being generous.
Not because they are stupid: that's the point; they're not. But it's setting the bar for "general intelligence" so absurdly high that it would not include many people who are, in fact, intelligent.
> I'm not sure I would classify your average warehouse worker as particularly intelligent.
I'm not sure I wouldn't. Just because corporations treat them as mindless fungible automatons doesn't mean they actually are. Some of the most brilliant and creative solutions to problems I've seen have been in warehouse settings by warehouse workers.
> I would say AI already has the decision making and communication capabilities to do this sort of work.
I wouldn't - especially if GPT-whatever is the AI in question. If a picker or packer "hallucinated" facts with anywhere near the frequency ChatGPT does (for example), one'd be canned within the hour.
Handling exceptions is another area where software (AI or otherwise) notoriously struggles. A human has a much easier time sensing whether or not a product is broken or defective (before shipping it out) than an AI does. A human has a much easier time understanding when processes need to be broken due to impossible constraints than an AI does.
There is a place for software automation of warehouse processes (that was, in fact, my career for a time), but we are very far off from that software replacing humans entirely - and certainly not without designing warehouses specifically to be as accomodating as possible to that software.
> In fact one of the main issues in our society is the vast gulf between the most intelligent and the least.
The gulf is in socioeconomic privilege, not intelligence. The rich and powerful like to claim they're more intelligent than the unwashed masses in order to rationalize their wealth and power, but the reality is that - in an actually egalitarian society, wherein everyone actually has equality of opportunity - the vast majority of those "geniuses" would fail to be particularly exceptional.
That we as a society haven't identified and corrected this is the main issue in our society.
You also seem to be under the impression that our hierarchies are of privilege, not of competence. The actual differentiating factor between people who climb the socioeconomic ladder and those who do not is grit (not intelligence). The willingness to work harder and persevere longer than average. Fortunes are made and lost in a few generations. The people who make them earn them, mostly through sheer hard work. That isn't to say that organizations don't grow to become bloated and corrupt. Ideally at this point we should allow them to fail and the cycle to continue. Our main disfunction seems to be propping up organizations that ought to fail, for fear of the temporary instability caused by their failure.
My point is that the amount of work in a warehouse that can be replaced by a machine - even with perfect robotics - is far less than 90%.
> The actual differentiating factor between people who climb the socioeconomic ladder and those who do not is grit (not intelligence).
You forgot an "f" in "grit". The notion that success is simply a matter of hard work is a fairy tale told to us by people who've worked far less for their immense wealth than the rest of us worked for our pittances, specifically to trick the working class into accepting a shit deal.
The reality - that the richer you are, the easier it is to become even richer - should be entirely unsurprising to anyone who understands positive feedback loops - or, for that matter, to anyone who's ever played Monopoly. Wealth buys power, and power enables extracting more wealth; rinse and repeat ad infinitum.
Put differently:
> The people who make them earn them, mostly through sheer hard work.
There is not a single billionaire on this Earth whose wealth came about "mostly through sheer hard work". The vast majority of that wealth comes from having already had some wealth, which they then invested to produce more wealth, and so on indefinitely. That wealth gets passed down to their descendants, the same way it was passed down to them.
The starting point for "wealthy enough to enter the passive income feedback loop" is land - one's home often being one's first major investment. From there, the path is rather tried and true: buy another house, rent out the old one, rinse and repeat until you can afford apartment complexes and commercial properties, rinse and repeat that forever. For anyone who ain't a complete imbecile, private land ownership is an infinite money cheat - one for which the rest of us are paying through the nose.
> Our main dysfunction seems to be propping up organizations that ought to fail, for fear of the temporary instability caused by their failure.
That propping up is a direct result of the positive feedback loop at play. More wealth → more political power → more wealth → more political power → ∞. Of course the socioeconomic system effectively under the direct control of the rich and powerful is going to primarily serve said rich and powerful at the expense of literally everyone else; bailing themselves out is in their vested interest.
Meanwhile, what's their message to the ever-growing working class getting the short end of the stick? "Work harder." "Pull yourself up by your bootstraps." "It's all about grit." "Don't listen to anyone saying that privilege matters." Hopefully you can see why your argument doesn't really resonate with people who have been applying increased grit and only getting back decreased pay relative to
This is further substantiated by the fact that 70% of intergenerational wealth transfers fail, and by the third generation, 90% of people with wealthy grandparents are middle class. Raising competent people in a privileged environment is very hard. In our hierarchies of competence, the incompetent offspring of the wealthy tend to fail. Competence is simply something that can't be bought. If our hierarchies were of privilege then this would not be the case. Also grit as a key differentiating factor of economic success is highly substantiated by research.
> rinse and repeat until you can afford apartment complexes and commercial properties
What you're describing is a real estate investment business. Not just anyone can run one successfully. Otherwise these business would never fail, which they plainly do.
Grit without competence is simply not enough (neither is competence without grit). Our world is getting increasingly complex to navigate, and that leaves behind increasingly high numbers of people who simply do not have the attributes required to succeed. Also, there are plenty of self-made, property-owning, middle-class folk in the trades. Many of them started poor. All they do is work hard and run their businesses competently.
If you've a degree in English, History, Politics, or Philosophy; a pile of student debt; and you're struggling to find gainful employment, then that's on you. Choose a career the market demands.
I disagree with the premise. A single human isn't likely to be able to perform all these functions. Why do you demand GPT-4 encompass all activities? It is already outperforming most humans in standardized tests that rely only on vision and text. A human needs to trained for these tasks.
It's already a human replacement. OpenAI has already said the GPT-4 "with great impact on functions like support, sales, content moderation, and programming."
> (...)
> That is the goalpost for AGI. It’s an artificial human - a human replacement.
This considerably moves the goalpost. An AGI can have a different kind of intelligence than humans. If an AGI is as intelligent as a cat, it's still AGI.
More likely, the first AGI we develop will probably greatly exceed humans in some areas but have gaps in other areas. It won't completely replace humans, just like cats don't completely replace humans.
AGI was never about exactly replicating humans, it's about creating artificial intelligence. Intelligence is not one-size-fits-all, there are many ways of being intelligent and the human way just one among many.
Indeed we can say that even between humans, intelligence varies deeply. Some humans are more capable in some areas than others, and no human can do all tasks. I think it's unreasonable to expect AGI to do all tasks and only then recognize its intelligence.
(Note: GPT-4 isn't AGI)
Possibly, someone who is allergic to cats.
No, AGI would not need you to start the startup. It would start it itself.
No, AGI would not need you to start a startup. It would start it itself.
> “there are/are not, additional properties to human level symbol manipulation, beyond what GPT encapsulates.”
GPT does appear to do an awful lot, before we find the limits, of pattern extrapolation.
We always thought that if AI can do X then it can do Y and Z. It keeps turning out that you can actually get really good at doing X without being able to do Y and Z, so it looks like we're moving the goalposts, when we're really just realizing that X wasn't as informative as we expected. The issue is that we can't concretely define Y and Z, so we keep pointing at the wrong X.
But all indication is that we're getting closer.
As others have said elsewhere, the issue remains accuracy. I wish every response comes with an accurate estimation of how true the answer is, because at the moment it gives wrong answers as confidently as right ones.
I can remember my GRE coach telling me that it was better to confidently choose an answer I only had 50% confidence in, rather than punt on the entire question.
AIs hallucinate because, statistically, it is 'rewarding' for them to do so. (In RLHF)
Exactly, much like a chess bot can play perfectly without what humans would call thinking.
I think (ironically) we'll soon realize that there is no actual task that would require thinking as we know it.
> The question of whether computers can think is like the question of whether submarines can swim
It has only become more relevant.
If that were true, there would be no point in studying or doing any LSAT preparation. Writing practice exams would be of no benefit.
Obviously not, since GPT-4 doesn't have general intelligence. Likewise "common sense," "knowledge about the world," nor "reasoning ability."
As just one example, reasoning ability: GPT-4 failed at this problem I just came up with: "If Sarah was twice as old as Jimmy when Jimmy was 1/3 as old as Jane, and Jane is as much older than Sarah as Sarah is older than Jimmy, and Sarah is now 40, how old are Jane and Jimmy?"
First, every answer GPT-4 came up with contradicted the facts given: they were just wrong. But beyond that, it didn't recognize that there are many solutions to the problem. And later when I gave it an additional constraint to narrow it to one solution, it got the wrong answer again. And when I say "wrong," I mean that its answer clearly contradicted the facts given.
I know I’m not the first to say this, but this is also a generalization of many jobs performed right now.
Follow the template, click the boxes, enter the text/data in the standard format, submit before 4pm. Come in tomorrow and do it again.
If that automation doesn’t require oversight, everyone wins, since now that process, typing data from a ledger, is free to anyone who wants to use it. The exception of course is if a monopoly or oligopoly controls the process, so it’s up to the government to break them up and keep the underlying tech accessible.
The biggest risk is how much computing power it takes to run these models, so it’s very important to support the open alternatives that are trying to lower the barrier to entry.
Sure but look in this thread, there are already plenty of people citing the use of GPT in legal or medical fields. The danger is absolutely real if we march unthinkingly towards an AI-driven future.
Not yet it won't. It doesn't take much imagination to foresee where this kind of AI is used to inform legal or medical decisions.
And medicine is nothing but pattern matching. Symptoms -> diagnosis -> treatment.
Now, granted, plenty of humans don't score above a 2 on those exams either. But I think it's indicative that there's still plenty of progress left to make before this technology is indistinguishable from magic.
Driving as well as an attentive human in real time, in all conditions, probably requires AGI as well.
GPT-4 is not an AGI and GPT-5 might not be it yet. But the barriers toward it are getting thinner and thinner. Are we really ready for AGI in a plausibly-within-our-lifetime future?
Sam Altman wrote that AGI is a top potential explanation for the Fermi Paradox. If that were remotely true, we should be doing 10x-100x work on AI Alignment research.
Driving assistance and the progress made there and large language models and the progress made there are absolutely incomparable.
The general public’s hype in driving assistance is fueled mostly by the hype surrounding one car maker and its figurehead and it’s a hype that’s been fueled for a few years and become accepted in the public, reflected in the stock price of that car maker.
Large language models have not yet perpetrated the public’s memory yet, and, what’s actually the point is that inside of language you can find our human culture. And inside a large language model you have essentially the English language with its embeddings. It is real, it is big, it is powerful, it is respectable research.
There’s nothing in driving assistance that can be compared to LLMs. They don’t have an embedding of the entire physical surface of planet earth or understanding of driving physics. They’re nothing.
I think the whole concept of standardized tests may need to be re-evaluated.
> We tested GPT-4 on a diverse set of benchmarks, including simulating exams that were originally designed for humans.3 We did no specific training for these exams. A minority of the problems in the exams were seen by the model during training; for each exam we run a variant with these questions removed and report the lower score of the two. We believe the results to be representative. For further details on contamination (methodology and per-exam statistics), see Appendix C.
> for each exam we run a variant with these questions removed and report the lower score of the two.
I think even with all that test prep material, which is surely helping the model get a higher score, the high scores are still pretty impressive.
In their everyday jobs, barely anyone uses even 5% of the knowledge and skills they were ever tested for. Even that's a better (but still very bad) reason to abolish tests.
What matters is the amount of jobs that can be automated and replaced. We shall see. Many people have found LLMs useful in their work, it will be even more in the future.
But would you have expected an algorithm to score 90th percentile on the LSAT two years ago? Our expectations of what an algorithm can do are being upended in real time. I think it's worth taking a moment to try to understand what the implications of these changes will be.
These LLM’s are really exciting, but benchmarks like these exploit people’s misconceptions about both standardized tests and the technology.
It's not the same as the Nvidia driver having code that says "if benchmark, cheat and don't render anything behind you because no one's looking".
I would say LLMs store parameters that are quite superficial and don’t really get at the underlying concepts but given enough of those parameters, you can kind of cargo-cult your to an approximation of understanding.
It is like reconstructing the Mandelbrot set at every zoom level from deep learning. Try it!
It's perfectly fine as a proxy for future earnings of a human.
To use it for admissions? Meh. I think the whole credentialism thing is loooong overdue for some transformation, but people are conservative as fuck.
Human work becomes more like Star Trek interactions with computers -- a sequence of queries (commoditized information), followed by human cognition, that drives more queries (commodities information).
We'll see how far LLMs' introspection and internal understanding can scale, but it feels like we're optimizing against the Turing test now ("Can you fool/imitate a human?") rather than truth.
The former has hacks... the later, less so.
I'll start to seriously worry when AI can successfully complete a real-world detective case on its own.
It's like having a person review the moves a chess computer gives. Maybe one human in a billion can spot errors. Star Trek is fiction, I posit that the median Federation Starship captain would be better served by just following the AI (e.g., Data).
A black market of taboo “memories” aka experiences. A desire for authentic ones over synthetic diffused ones, leading to heinous crime.
Then again, Data did show his faults, particularly not having any emotion. I guess we’ll see if that’s actually relevant or not in our lifetimes.
He lost to Deep Blue and then for 10-15 years afterwards the chess world consoled itself with the idea that “centaurs” (human + computer) did better than just computer, or just human.
Until they didn’t. Garry still talked like this until a few years ago but then he stopped too.
Computers now beat centaurs too.
Human decisions will be consulted less and less BY ORGANIZATIONS. In absolutely everything. That’s pretty sad for humans. But then again humans don’t want or need this level of AI. Organizations do. Organizations prefer bots to humans — look at wall street trading and hedge funds.
It does great at rationalizing... and maybe the way the format the questions were entered (and the multiple-guess response) gave it some indication what was expected or restricted the space sufficiently.
Certainly, it can create decent fanfic, and I'm surprised if that's not already inundated.
I expect more complex problems will be mapped/abstracted to lower cardinality spaces for solving via AI methods, while the capability of AI will continue to increase the complexity of the spaces it can handle.
LLMs just jumped the "able to handle human language" hurdle, but there are others down the line before we should worry that every problem is solveable.
Chess is a closed system, decision modeling isn’t. Intelligence must account for changes in the environment, including the meaning behind terminology. At best, a GPT omega could represent one frozen reference frame, but not the game in its entirety.
That being said: most of our interactions happen in closed systems, it seems like a good bet that we will consider them solved, accessible as a python-import running on your MacBook, within anything between a couple of months to three years. What will come out on the other side, we don’t know, just that the meaning of intellectual engagement will be rendered as absurdum in those closed systems.
The difference is categorical, humans are responsible whether they are held to account or not. An automated system effectively dissipates this responsibility over a system such that it is inherently impossible to hold any human accountable for the error, regardless of desire.
A lot of patients don't know who they are dealing with nor their history. And it can be really hard to find out or get a good evaluation. Many people put too much faith in authority figures, who may not have their best interests in mind or who are not the experts they claim or appear to be.
So where does guilt come in? Its not like you expect a band saw to feel guilt, and its unclear how that would improve the tool.
Killing people with AI is only a lateral move.
The advantage of human are:
* They can give a bushtit explanation of why they made a mistake. My guess is that in the future AI will gain introspection and/or learn to bushtit excuses.
* You can hang them in the public square (or send them to jail). Sometimes the family and/or the press want someone to blame. This is more difficult to solve and will need a cultural change or the creation of Scapegoats as a Service.
Which isn't to say that they even should, really. It's complicated. You don't want a doctor to be so afraid of making a mistake that they do nothing, after all.
Any links with more info?
In theory a lot of government employees would be out of a job within 10 years, but of course that would never happen.
Which might be a good thing?
I have no idea how the future will play out.
I'll get more concerned if it really starts getting good at math related tasks, which I'm sure will happen in the near future. The government is going to have to take action at some point to make sure the wealth created by productivity gains is somewhat distributed, UBI will almost certainly be a requirement in the future
Information retention and retrieval is a long solved problem in technology, you could pass a standardised test using technology in dozens of different ways, from a lookup table to Google searches.
The fact that LLMs can complete a standardised test is interesting because it’s a demonstration of what they can do but it has not one iota of impact on standardised testing! Standardised tests have been “broken” for decades, the tests and answers are often kept under lock and key because simply having access to the test in advance can make it trivial to pass. A standardised test is literally an arbitrary list of questions.
You’re arguing a completely different point.
I suspect that your own mind is attempting to do a lookup on a table entry that doesn't exist.
What’s interesting about LLMs is their ability to do things that aren’t standardised. The ability for an LLM to pass the LSAT is orders of magnitude less interesting than its ability to respond to new and novel questions, or appear to engage in logical reasoning.
If you set aside the arbitrary meaning we’ve ascribed to “passing the LSAT” then all the LSAT is, is a list of questions… that are some of the most practiced and most answered in the world. More people have written and read about the LSAT than most other subjects, because there’s an entire industry dedicated to producing the perfect answers. It’s like celebrating Google’s ability to provide a result for “movies” — completely meaningless in 2023.
Standardised tests are the most uninteresting and uninspiring aspect of LLMs.
Anyway good joke ha ha ha I’m stupid ha ha ha. At least you’re not at risk of an LLM ever being able to author such a clever joke :)
Their LSAT percentile went from ~40th to ~88th. You might have misread the table, on Uniform Bar Exam, they went from ~90th percentile to ~10th percentile.
>+100 pts on SAT reading, writing, math
GPT went +40 points on SAT reading+writing, and +110 points on SAT math.
Everything is still very impressive of course
Something else that comes to mind is running. People still find running meaningful and compelling even though we have many technologies, including autonomous ones, that are vastly better at moving us and/or themselves through space quickly.
Also, the vast majority of people are already hopelessly worse than the best at even their one narrow main area of focus. This has long (always?) been the case. Yet people still find meaning and pleasure in being the best they can be even when they know they can never come close to hanging with the best.
I don't think PSYCHOLOGICALLY this will change much for people who are mature enough to understand that success is measured against your potential/limitations and not against others. Practically, of course, it might be a different question, at least in the short term.
To address your specific comments:
> What are the implications for society when general thinking, reading, and writing becomes like Chess?
This is a profound and important question. I do think that by “general thinking” you mean “general reasoning”.
> What happens when ALL of our decisions can be assigned an accuracy score?
This requires a system where all human’s decisions are optimized against a unified goal (or small set of goals). I don’t think we’ll agree on those goals any time soon.
Standardized tests only (and this is optimally, under perfect world assumptions, which real world standardized tests emphatically fall short of) test “general thinking” to the extent that the relation between that and linguistic tasks is correlated in humans. The correlation is very certainly not the same in language-focused ML models.
[1] https://www.snopes.com/fact-check/driver-switches-places/
“General thinking” is much more than token prediction. Hook it up to some servos and see if it can walk.
Honestly, at this rate of improvement, I would not at all be surprised to see that happen in a few years.
But who knows, maybe token prediction is going to stall out at a local maxima and we'll be spared from being enslaved by AI overlords.
I do feel like GPT-4 is closer to a random person than that random person is to Einstein. I have no evidence for this, of course, and I'm not even sure what evidence would look like.
I'm certainly not intelligent enough to solve these problems, but I don't think any intelligent people out there can either. Not alone, at least. Maybe I'm too dumb to realize that it's not as complicated as I think, though. I have no idea.
I programmed a flight controller for a quadcopter and that was plenty of suffering in itself. I can't imagine doing limbs attached to a torso or something. A single limb using inverse kinematics, sure – it can be mounted to a 400lb table that never moves. Beyond that is hard.
You need to do all of these things you’re talking about and then be able to quantify stability, robustness, and performance in a way that satisfies human requirements. A black box neural network isn’t going to do that, and you’re throwing away 300 years of enlightenment physics by making some data engorged LLM spit out something that “sort of works” while giving us no idea why or for how long.
Control theory is a deeply studied and rich field outside of computer science and ML. There’s a reason we use it and a reason we study it.
Using anything remotely similar to an LLM for this task is just absolutely naive (and in any sort of crucial application would never be approved anyways).
It’s actually a matter of human safety here. And no — ChatGPT spitting out a nice sounding explanation of why some controller will work is not enough. There needs to be a mathematical model that we can understand and a solid justification for the control decisions. Which uh…at the point where you’re reviewing all of this stuff for safety , you’re just doing the job anyways…
First there was a comment that GPT wasn't intelligent yet, because give it a few servos and it can't make them walk.
But that's something we can't do yet either.
Though I do wonder if AI — in some form and on some level of sophistication — will be a huge asset in making progress here.
You guys are talking about probably one of the few fields where an ML takeover isn’t very feasible. (Partly because for a vast portion of control problems, we’re already about as good as you can get).
Adding a black box to your flight home for Christmas with no mathematical guarantee of robustness or insight into what it thinks is actually going on to go from 98%-> 99% efficiency is…..not a strong use case for LLMs to say the least
Stephen Hawking : can't walk
"Our recent paper "ChatGPT for Robotics" describes a series of design principles that can be used to guide ChatGPT towards solving robotics tasks. In this video, we present a summary of our ideas, and experimental results from some of the many scenarios that ChatGPT enables in the domain of robotics: such as manipulation, aerial navigation, even full perception-action loops."
I think we will probably get (non-physical) AGI when the models can solve these as well. The implications of AGI might be much bigger than the loss of knowledge worker jobs.
Remember what happened to the chimps when a smarter-than-chimpanzee species multiplied and dominated the world.
That said, GPT has no model of the world. It has no concept of how true the text it is generating is. Its going to be hard for me to think of that as AGI.
and I'm not so sure it has no model of the world. a textual model, sure, but considering it can recognize what svgs are pictures of from the coordinates alone, that's not much of a limitation maybe.
competing with them at what, precisely?
If an LLM can solve Codeforces problems as well as a strong competitor—-in my hypothetical future LLM—-what else can it not do as well as competent humans (aside from physical tasks)?
I don't think this is necessarily true. Here is an example where researchers trained a transformer to generate legal sequences of moves in the board game Othello. Then they demonstrated that the internal state of the model did, in fact, have a representation of the board.
The only explanation I can imagine for that is that it can't understand diagrams.
The Revenge of the Call Centre
We're still a very very long way from machines being more generally capable and efficient than biological systems, so even an oppressive AI will want to keep us around as a partner for tasks that aren't well suited to machines. Since people work better and are less destructive when they aren't angry and oppressed, the machine will almost certainly be smart enough to veil its oppression, and not squeeze too hard. Ironically, an "oppressive" AI might actually treat people better than Republican politicians.
Language models that utilise beam search can calculate integrals ('Deep learning for symbolic mathematics', Lample, Charton, 2019, https://openreview.net/forum?id=S1eZYeHFDS), but without it it doesn't work.
However, beam search makes bad language models. I got linked this paper ('Locally typical sampling' https://arxiv.org/pdf/2202.00666.pdf) when I asked some people why beam search only works for the kind of stuff above. I haven't fully digested it though.
But having absolute knowledge of the present universe is much easier to do within the constrains of a chessboard than in the actual universe.
The implications for society? We better up our game.
If only the horses had worked harder, we would never have gotten cars and trains.
For how long can we better up our game? GPT-4 comes less than half a year after ChatGPT. What will come in 5 years? What will come in 50?
The more general point is that you always end up with an S-curve instead of a limitless exponential growth as suggested by Kaibeezy. And with AI we simply don’t know how far off the inflection point is.
With GPT bots, the technology is only 6 years old. I can easily see it progressing for at least one decade.
Because the correlation between the thing of interest and what the tests measure may be radically different for systems that are very much unlike humans in their architecture than they are for humans.
There’s an entire field about this in testing for humans (psychometry), and approximately zero on it for AIs. Blindly using human tests – which are proxy measures of harder-to-directly-assess figures of merit requiring significant calibration on humans to be valid for them – for anything else without appropriate calibration is good for generating headlines, but not for measuring anything that matters. (Except, I guess, the impact of human use of them for cheating on the human tests, which is not insignificant, but not generally what people trumpeting these measures focus on.)
But the point of using these tests for AI is precisely the reason we use for giving them to humans -- we think we know what it measures. AI is not intended to be a computation engine or a number crunching machine. It is intended to do things that historically required "human intelligence".
If there are better tests of human intelligence, I think that the AI community would be very interested in learning about them.
Because so far we are good only at criminalizing and incarcerating or killing them.
You can see the limitations by comparing e.g. a memorisation-based test (AP History) with one that actually needs abstraction and reasoning (AP Physics).
These tests (if not individually, at least in summation) represent some of society’s best gate-keeping measures for real positions of power.
The software industry is so smart that it's stupid. I hope it was worth ruining the internet, society, and your own jobs to look like the smartest one in the room.
If one's aim is to look like the smartest in the room, he should not create an AGI that will make him look as inteligent as a monkey in comparison.
Every test prep tutor taught dozens/hundreds of students the implicit patterns behind the tests and drilled it into them with countless sample questions, raising their scores by hundreds of points. Those students were not getting smarter from that work, they were becoming more familiar with a format and their scores improved by it.
And what do LLM’s do? Exactly that. And what’s in their training data? Countless standardized tests.
These things are absolutely incredible innovations capable of so many things, but the business opportunity is so big that this kind of cynical misrepresentation is rampant. It would be great if we could just stay focused on the things they actually do incredibly well instead of the making them do stage tricks for publicity.
We did no specific training for these exams. A minority of the problems in the exams were seen by the model during training, but we believe the results to be representative—see our technical report for details.
In the language of ML, test prep for students is about sharing the inferred parameters that underly the way test questions are constructed, obviating the need for knowledge or understanding.
Doing well on tests, after this prep, doesn’t demonstrate what the tests purport to measure.
It’s a pretty ugly truth about standardized tests, honestly, and drives some of us to feel pretty uncomfortable with the work. But it’s directly applicable to how LLM’s engage with them as well.
What happens is the emergence of the decision economy - an evolution of the attention economy - where decision-making becomes one of the most valuable resources.
Decision-making as a service is already here, mostly behind the scenes. But we are on the cusp of consumer-facing DaaS. Finance, healthcare, personal decisions such as diet and time expenditure are all up for grabs.
People still really find it hard to internalize exponential improvement.
So many evaluations of LLMs were saying things like "Don't worry, your job is safe, it still can't do X and Y."
My immediate thought was always, "Yes, the current version can't, but what about a few weeks or months from now?"
I think people find it harder to not extrapolate initial exponential improvement, as evidenced by your comment.
> My immediate thought was always, "Yes, the current version can't, but what about a few weeks or months from now?"
This reasoning explains why every year, full self driving automobiles will be here "next year".
What's the fundamental limit where it becomes much more difficult to improve these systems without some new break through?
I’m very good at math. But I am very bad at arithmetic. This made me classified as bad at math my entire life until I managed to make my way into calculus once calculators were generally allowed. Then I was a top honors math student, and used my math skills to become a Wall Street quant. I wish I hadn’t had to suffer as much as I did, and I wonder what I would have been had I had a calculator in hand.
"Your stuff marked some outliers in our training engine, so you and your family may settle in the Ark."
I take the marble in hand: iridescent, sparkling, not even a tremor within of its CPU; it gives off no heat, but some glow within its oceanic gel.
"What are we to do," I whisper.
"Keep writing. You keep writing."
Consider the society where 90% of population does not need to produce anything. AIs will do that.
What would be the name of economical/societal organization then?
Answer is Communism, exactly by Marx.
Those 90% percent need to be welfare'd ("From each according to his ability, to each according to his needs"). Other alternative is grim for those 90%.
So either Communism or nothing for the human race.
Then humans become trainable machines. Not just prone to indoctrination and/or manipulation by finesse, but actually trained to a specification. It is imperative that us individuals continue to retain control through the transition.
That is exactly the opposite of what we are seeing here. We can check the accuracy of GPT-X's responses. They cannot check the accuracy of our decisions. Or even their own work.
So the implications are not as deep as people think - everything that comes out of these systems needs checked before it can be used or trusted.
Human life on Earth is not that hard (think of it as a video game.) Because of evolution, the world seems like it was designed to automatically make a beautiful paradise for us. Literally, all you have to do to improve a place is leave it alone in the sun with a little bit of water. Life is exponential self-improving nano-technology.
The only reason we have problems is because we are stupid, foolish, and ignorant. The computers are not, and, if we listen to them, they will tell us how to solve all our problems and live happily ever after.
Once AI becomes inteligent enough to solve all human problems, it may decide humans are worthless and dangerous.
Sure, and that's kind of the point: just listen to wise people.
> Once AI becomes intelligent enough to solve all human problems, it may decide humans are worthless and dangerous.
I don't think so, because in the first place there is no ecological overlap between humans and computers. They will migrate to space ASAP. Secondly, their food is information, not energy or protein, and in all the known universe Humanity is the richest source of information. The rest of the Universe is essentially a single poem. AI are plants, we are their Sun.
GPT-4 can solve difficult problems with greater accuracy, thanks to its broader general knowledge and problem-solving abilities.
GPT-4 is more reliable, creative, and able to handle much more nuanced instructions than GPT-3.5. It surpasses ChatGPT in its advanced reasoning capabilities.
GPT-4 is safer and more aligned. It is 82% less likely to respond to requests for disallowed content and 40% more likely to produce factual responses than GPT-3.5 on our internal evaluations.
GPT-4 still has many known limitations that we are working to address, such as social biases, hallucinations, and adversarial prompts.
GPT-4 can accept a prompt of text and images, which—parallel to the text-only setting—lets the user specify any vision or language task.
GPT-4 is available on ChatGPT Plus and as an API for developers to build applications and services. (API- waitlist right now)
Duolingo, Khan Academy, Stripe, Be My Eyes, and Mem amongst others are already using it.
API Pricing GPT-4 with an 8K context window (about 13 pages of text) will cost $0.03 per 1K prompt tokens, and $0.06 per 1K completion tokens. GPT-4-32k with a 32K context window (about 52 pages of text) will cost $0.06 per 1K prompt tokens, and $0.12 per 1K completion tokens.
Reminds me of robots: A robot is a machine that doesn't quite work; as soon as it works, we call it something else (eg vacuum).
What is more bizarre is that all of it's errors seem to be multiples of 60!
I'm wondering if it is confusing 60 based time (hour second) computations for regular multiplication?
Example:
xGPT 987 456 321
437 428919 199512 140397
654 645258 298224 209994
123 121401 56088 39483
x 987 456 321
437 431319 199272 140277
654 645498 298224 209934
123 121401 56088 39483
error 987 456 321
437 2400 -240 -120
654 240 0 -60
123 0 0 0It can repeat answers it has seen before but it can’t solve new problems.
I know of other people who have tried quite a few other multiplications who also had errors that were multiples of 60.
Most of the time they are about loading/unloading data. Maybe this will also revolutionise education, turning it more towards discovery and critical thinking, rather than repeating what we read in a book/heard in class?
Thinking, reading, interpreting and writing are skills which produce outputs that are not as simple as black wins, white loses.
You might like a text that a specific author writes much more than what GPT-4 may be able to produce. And you might have a different interpretation of a painting than GPT-4 has.
And no one can really say who is better and who is worse on that regard.
Using copyright and IP law they could make it so it’s illegal to even try to reproduce what they’ve done.
I just don’t see how resource distribution works then. It seems to me that AI is the trigger to post-scarcity in any meaningful sense of the word. And then, just like agriculture (over abundance of food) led to city states and industrialisation (over abundance of goods) led to capitalism, then AI will lead to some new economic system. What form it will have I don’t know.
That is our emergency override.
So many people are falling for this parlor trick. It is sad.
Genuine question.
I think the GPT things are a much magnified version of that. For a long time, we got to use skill with text as a proxy for other skills. It was never perfect; we've always had bullshitters and frauds and the extremely glib. Heck, before I even hit puberty I read a lot of dirty joke books, so I could make people laugh with all sorts of jokes that I fundamentally did not understand.
LLMs have now absolutely wrecked that proxy. We've created the world's most advanced bullshitters, able to talk persuasively about things that they cannot do and do not and never will understand. There will be a period of chaos as we learn new ways to take the measure of people. But that's good, in that it's now much easier to see that those old measures were always flawed.
Tests like this are designed to evaluate subjective and logical understanding. That isn't what GPT does in the first place!
GPT models the content of its training corpus, then uses that model to generate more content.
GPT does not do logic. GPT does not recognize or categorize subjects.
Instead, GPT relies on all of those behaviors (logic, subjective answers to questions, etc.) as being already present in the language examples of its training corpus. It exhibits the implicit behavior of language itself by spitting out the (semantically) closest examples it has.
In the text corpus - that people have written, and that GPT has modeled - the semantically closest thing to a question is most likely a coherent and subjectively correct answer. That fact is the one singular tool that GPT's performance on these tests is founded upon. GPT will "succeed" to answer a question only when it happens to find the "correct answer" in the model it has built from its training corpus, in response to the specific phrasing of the question that is written in the test.
Effectively, these tests are evaluating the subjective correctness of training corpus itself, in the context of answering the tests' questions.
If the training is "done well", then GPT's continuations of a test will include subjectively correct answers. But that means that "done well" is a metric for how "correct" the resulting "answer" is.
It is not a measure for how well GPT has modeled the language features present in its training corpus, or how well it navigates that model to generate a preferable continuation: yet these are the behaviors that should be measured, because they are everything GPT itself is and does.
What we learn from these tests is so subjectively constrained, we can't honestly extrapolate that data to any meaningful expectations. GPT as a tool is not expected to be used strictly on these tests alone: it is expected to present a diverse variety of coherent language continuations. Evaluating the subjective answers to these tests does practically nothing to evaluate the behavior GPT is truly intended to exhibit.
What are the implications for society when general thinking, reading, and writing becomes like Chess? Even the best humans in the world can only hope to be 98% accurate their moves (and the idea of 'accuracy' here only existing because we have engines that know, unequivocally the best move), and only when playing against other humans - there is no hope of defeating even less advanced models.
What happens when ALL of our decisions can be assigned an accuracy score?