And whether or not it actually is feasible, many are betting millions that it is, and marketing it as possible to keep the innovation machine running until we achieve it
> ... I always say about [a language model], the [linguistic] appearance makes a promise about what it can do. [Clippy] was this little [cartoon paper clip]. It didn’t promise much—you saw it and thought, that’s not going to [write the next great novel]. But you can imagine it [offering limited help]. But [human language] sort of promises it can [write] anything a human can. And that’s why it’s so attractive to people—it’s selling a promise that is amazing.
The difference between the promise and reality of LLMs and the difference between the promise and reality of humanoid robots are a different order of magnitude.
When a language model fumbles, its mistakes are still wrapped in convincing writing, so the error is only apparent if the user already knows what the answer should be.
When a humanoid robot fumbles, its mistakes are obvious because the physical world offers immediate feedback.
It's the difference between lying on your résumé that you're a world-class gymnast, and having to actually perform.
How much of this is due to nearly all humans already having advanced knowledge of what they would expect out of a humanoid robot in the home?
With the gymnast example, as a non-gymnast, I don’t know the difference between a high and low scoring routine on the floor or beam. If a humanoid robot did a routine and didn’t fall, I would assume all is well. I don’t know the technical details of what is required for a gymnastics competition.
This seems like the same idea as an LLM writing a paper that looks correct to someone who doesn’t already know the answer.
In a home context, this could look like the robot not practicing proper food safety or storage around someone who doesn’t know the details about that kind of thing, which is a good number of people. What it’s doing might look correct enough, and it produces food you can eat… all is well, until you get sick and don’t know why.
Which gymnast competition? The well known ones are more beauty contests with/on gymnastics equipment. However there are also competitions where they measure objective things. I know what I like to see in a beauty contest, but that is also subjective. I too don't know what a technical competition is measuring, but I know they have objective things they look for.
I don’t know what you’re referring to when talking about gymnastics as a beauty contest.
I’m not an expert, but I know there are specific moves with various degrees of difficulty. I believe there is a max score based on that difficulty level, and any imperfection will lower that score, such as a foot pointed or flexed the wrong way at the wrong time, taking an extra step on a landing, etc.
I know all these rules exist, but I’m not an expert where I can say someone had their foot flexed when it should have been pointed. These details would go over my head, where a humanoid robot might get a pass from me, while an actual gymnast or judge would be able to see faults.
I’m kinda torn between “genAI powered robots will have a ground truth reality as a reference, so they will ultimately be more grounded and effective that LLMs” and “LLMs are like drunk uncle Steve with his PHD swimming in vodka, and using genAI in robots will end up as well as having drunk uncle Steve drive home”.
Guardrails on tasks it will attempt are inevitable, but I can also see that becoming a paywalled enshitification farm.
Yeah, imagine "that guy from the pub" who is unemployed for years because he claims to be "overqualified for everything", and then add that he knows exactly how to convince you that he is capable of EVERYTHING you throw at him...
LLMs are much closer in practice, they're already useful for a pretty wide range of tasks. Humanoid robots are still comically clumsy and limited, barely able to complete scripted tech demos.
The difference is very easy to define and notorious difficult to solve: it is physics. And man is physics a hard problem to "solve".
Welcome to the world of hard tech not easy machine learning models. Capital is in short supply, it doesn't go nearly as far and you don't get wild multiples in return if you even get any.
I cannot quite tell what point this is trying to make. LLMs are just the next Clippy? As far as I can remember no one actually liked Clippy, so my read is you are not a fan of LLMs, but I could see it going either way.
I took it to mean that the way LLM's use natural language causes the typical observer to feel as if they can perform far more than what they actually can. Akin to the analogy of humanoid robots.
It plays off of the "if it looks like a duck, quacks like a duck, and walks like a duck" idiom, which of course isn't foolproof and gives avenue to the kind of spectacular advertising that is fueling this hype.
My mom was lamenting car insurance quotes, so I told her to ask AI. She did, then had it do a Monte Carlo simulation for all the insurances she the AI felt she was qualified for.
It happily replied that it did 1 million monte carlo simulations and here was the result.
To this day I don't think she fundamentally groks that LLMs cannot calculate.
For me, it was a friend that was wildly impressed by ChatGPT (before it could search the web) had "analyzed recent market news and stock price graphs" to give him stock recommendations.
I used Gemini to take "0.3 grams of KNO3 will raise the nitrate level of 10 gallons of water 4.84 ppm" and generate tables of how many grams of dry fertilizer for 1ppm, 5ppm, 10ppm for my planted aquariums of 144 and 3000 liters. It calculated them perfectly.
ChatGPT can easily do Monte Carlo simulation in its "thinking" step, and has done many times for me. e.g. I asked it to compare savings interest between regular banks and median returns from premium bonds. It's not difficult at all for it to do, you can see the code it's generated to do it + the output, easy to inspect
I understood it as mocking the iRobot's founder quote, that what he says is a false promise, could just as well be applied to LLMs, where it has been a true promise (but he says the opposite mockingly).
I would say the same delusion even applies to the field of machine learning in general.
The "API" of trainable algorithms is essentially "arbitrary bunch of data in -> thing I want out" and the magic algorithm in the middle will figure out the rest.
Because "thing I want" is given as a list of examples, you're not even required to come up with a clear definition of what it exactly is that you want. In fact, it's a major "selling point" of the field that you don't have to.
But all of that creates the illusion that machine learning / "AI" would be able to generate a robust algorithm for any correspondence, as long as you can package up a trainset with enough examples and shore up enough "compute" to do the number crunching. Predict intelligence from passport photos? Or chances of campaign success from political speeches? No problem! Economic outlook from tea leaves? Sure thing!
The setup will not even tell you if your idea just needs more tweaks or fundamentally can't work. In both cases, all you get is a less-than-ideal number in your chosen evaluation metric.
The process is definitely vulnerable to magical thinking.
I think it is possible to avoid, though, by asking if humans can be generally good at the task in question, if working through the implied interface restrictions, and then evaluating whether the required skills can be reflected in an available training data set.
If either of those cannot be definitively answered, it’s probably not going to work.
An interesting example here is the failure of self driving vehicles based on image sensors.
My take is that most of the problems are because a significant fraction of the actual required training data is poorly represented in data that can be collected from driving experiences alone.
As in: If you want a car to be able to drive safely around humans, you need to understand a lot about what humans do and think about. - then apply that same requirement to everything else that occasionally appears in the operational environment.
To understand some traffic management strategies expressed in infrastructure, you’ll need to understand, to some degree, the goals of the traffic management strategy, aka “what were they thinking when they made this intersection?”.
It’s not all stuff you can magically gather from dashcams.
Yeah, my understanding was also that the (remaining) hard part of self-driving cars is guessing the intentions of other traffic participants. There are a lot of assumptions human drivers can make about pedestrians, e.g. whether a pedestrian has seen the car or not, whether they will wait for it, have no intention of crossing at all - or will just run across the street.
A model might potentially be able to understand those situations, but it would need a lot of highly task specific training data and it would never be clear if the training really covered all possible situations.
The other problem I see is that a lot of situations in traffic are really two-way communication, even if it is nonverbal and sometimes so implicit we don't realize it. But usually pedestrians will also try to infer what the driver is thinking whether he saw them, etc.
In those situations, a self-driving car is simply a fundamentally different kind of traffic participant and pedestrians will interact with it differently than they would with a normal car. That problem is independent of machine learning and seems much harder to solve to me.
That's not the "API" that's powered the AI boom though. What you're talking about is supervised learning. Generative AI is mostly unsupervised. It's "bunch of data -> similar data conditioned on some input". This goalless nature is one of its strengths.
The sort of questions you're talking about are primarily popular in academia. Run some MLRs against some random dataset you found, publish a paper, maybe do a press release and sell a story to some gullible journalist. It doesn't have huge value. But generative AI isn't like that.
Hasnt written a great novel, wont ever write a great novel, will definitely write regurgitated slop that midwit tech slaves steeped in the works of Malcolm Gladwell and Co. will read four words of and proclaim "Dostoevsky!"
I think that the main reason that LLM writing fails so badly in a field one might assume it would excel in is the lack of being able to model a theory of mind for the reader.
While I have seen LLMs produce some ham-fisted attempts at manipulating the state of mind of the reader, I think that the human process is so obfuscated that it only shows up in occasional echoes and shadows in the training set.
It might be possible to develop a training set that reflected perception and internal mental state vs input using (magic brain scan technology) that could change this, but right now the emotional state of the reader is just missing from the training data.
This take is so tiring, here's one of the most surprising things we've ever invented, and people are going "IT CAN'T WRITE DOSTOEVSKY". It's fine if y'all are so jaded, but can you at least keep it to yourselves?
LOL it's awesome, amazing tool, and I never saw it coming glad to have it, I'm responding to GP, what it does is nothing like good writing at all, and the only people who think otherwise are without exception people that have little to no exposure or training in any human arts.
When humans do writing, the quality improves by refining multiple drafts, making sketches and notes of the characters and situations and so on, before synthesizing the final text. A lot of preparation and thought goes into it.
If you just ask an LLM to write something off the cuff, it'll be bad. But doing a lot of prep with a human author guiding it? Not Dostoyevsky level, but not pure slop.
Creative writing sites without anti ai rules are getting swamped. Disheartening looking for people to read and provide feedback when hundreds of users churn out 40k word stories on a daily basis.
piggybacking on everything you said, which is all true: Code is not a science, despite what pedants would have you believe. The annoying answer to "what's correct" code is, "it depends." Code is just a tool used to achieve a goal.
why do you need to debug it?
or, why do you need to debug it?
my crazy thoughts, which I am open to being wrong about:
this is still the old way of thinking. generative AI is a probability function—the results exist in a probability space. we expect to find a result at a specific point in that space, but we don't know the inputs to achieve it until it is solved. catch 22.
instead, we must embrace the chaos by building attractors in the system, and pruning results that stray too far from the boundaries we set. we should focus only on macro-level results, not the code. multiple solutions existing at any given time. if anything is "debugging", it is a generative AI agent.
I guess a lot of us are trying this (naturally) as solo devs, where we can take an engineering-first mindset and build a machine or factory that spits out gizmos. I haven't gotten to the finish line, mostly because for me, the holy grail is code confidence via e2e tests that the agent generated (separately, not alongside the implementation).
This presupposes that human value only exists in the things current AI tech can replace—pattern recognition/creation. I'd wager the same argument was made when hand-crafted things were being replaced with industrialized products.
I'm not saying those things aren't valuable, or that humans can't express social and spiritual value in those ways, but that human value doesn't only exist there. And so, to give AI the power of complete dehumanization is to reduce humans to just pattern followers. I don't believe that is the case.
> I'm not saying those things aren't valuable, or that humans can't express social and spiritual value in those ways, but that human value doesn't only exist there.
This sounds sort of like a "God of the gaps" argument.
Yes, we could say that humanity is left to express itself in the margins between the things machines have automated away. As automation increases its capabilities, we just wander around looking for some untouched back-alley or dark corner the robots haven't swept through yet and do our dancing and poetry slams there until the machines arrive forcing us to again scurry away.
But at that point, who is the master, us or the machines?
What we still get paid to do is different than what we're still able to do. I'm still able to knit a sweater if I find it enjoyable. Some folks can even do it for a living (but maybe not a living wage)
If this came to pass, the population would be stripped of dignity pretty much en masse. We need to feel competent, useful, and connected to people. If people feel they have nothing left, then their response will be extremely ugly.
> And so, to give AI the power of complete dehumanization is to reduce humans to just pattern followers.
It would but I don't think that's what they're saying. The agent of dehumanization isn't the technology, but the selection of what the technology is applied to. Or like the quip "we made an AI that creates, freeing up more time for you to work."
Wherever human value, however you define that, exists or is created by people, what does it look like to apply this technology such that human value increases? Does that look like how we're applying it? The article seems to me to be much more focused on how this is actually being used right now rather than how it could be.
We should actively encourage fluidity in purpose, too much rigidity or militant clinging to ideas is insecurity or attempts at absolving personal responsibility.
Resilience and strength in our civilisation comes from confidence in our competence,
not sanctifying patterns so we don’t have to think.
We need to encourage and support fluidity, domain knowledge is commoditised, the future is fluid composition.
Great, tell that to someone who spent years honing their skills that it's too bad the rug was pulled out from beneath you, time to start over from the bottom again.
Maybe there would be merit to this notion if society provided the necessary safety net for this person to start over.
Agreed, I think there should be much more safety net for people to start over and be more fluid, I definitly think the weird "Full time employed or homeless" thing has to change
> We should actively encourage fluidity in purpose
I don't think we should assume most people are capable of what you describe. Assigning "should" to this assumes what you're describing is psychologically tenable across a large population.
> too much rigidity or militant clinging to ideas is insecurity or attempts at absolving personal responsibility.
Or maybe some people have a singular focus in life and that's ok. And maybe we should be talking about the responsibility of the companies exploiting everyone's content to create these models, or the responsibility of government to provide relief and transition planning for people impacted, etc.
To frame this as a personal responsibility issue seems fairly disconnected from the reality that most people face. For most people, AI is something that is happening to them, not something they are responsible for.
And to whatever extent we each do have personal responsibility for our careers, this does not negate the incoming harms currently unfolding.
Due to how AI works its only a matter of time till its better at pretty much everything humans do beside “living”.
People tend to talk about any AI related topic comparing it to any industrial shift that happened in the past.
But its much Much MUCH bigger this time. Mostly because AI can make itself better, it will be better and it is better with every passing month.
Its a matter of years until it can completely replace humans in any form of intellectual work.
And those are not mine words but smartest ppl in the world, like AI grandfather.
We humans think we are special. That there wont be something better than us. But we are in the middle of the process of creating something better.
It will be better. Smarter. Not tired. Wont be sick. Wont ever complain.
And it IS ALREADY and WILL replace a lot of jobs and it will not create new ones purely due to efficiency gains and lack of brainpower in majority of ppl who will be laid off.
Not everyone is a noble prize winner. And soon we will need only such ppl to advance AI.
Can it? I'm pretty sure current AI (not just LLMs, but neural nets more generally) require human feedback to prevent overfitting. Fundamentally eschewing any fear or hope of the singularity as predicted.
AI can not make itself better because it can not meaningfully define what better means.
> Its a matter of years until it can completely replace humans in any form of intellectual work.
This is sensationalism. There’s no evidence in favor of it. LLMs are useful in small, specific contexts with many guardrails and heavy supervision. Without human-generated prior art for that context they’re effectively useless. There’s no reason to believe that the current technical path will lead to much better than this.
To say that this is pure fantasy when there are more and more demos of humanoid robots doing menial tasks, and the costs of those robots are coming down is ... well something. Anger, denial (you are here)...
To say that this is pure fantasy when there are more and more demos of humanoid robots doing menial tasks
A demo is one thing. Being deployed in the real world is something else.
The only thing I've seen humanoid robots doing is dancing and occasionally a backflip or two. And even most of that is with human control.
The only menial task I ever saw a humanoid robot do so far is to take bags off of a conveyor belt, flatten them out and put them on another belt. It did it at about 1/10th the speed of a human, and some still ended up on the floor. This was about a month ago, so the state of the art is still in the demo stage.
I'm waiting. You're talking to someone who believed that self-driving vehicles would put truckers out of work in a decade right around 2012. I didn't think that one through. The world is very complicated and human beings are the cheapest and most effective way to get physical things done.
The risk raised in the article is that AI is being promoted beyond its scope (pattern recognition/creation) to legal/moral choice determination.
The techo-optimists will claim that legal/moral choices may be nothing more than the sum of various pattern-recognition mechanisms...
My take on the article is that this is missing a deep point: AI cannot have a human-centered morality/legality because it can never be human. It can only ever amplify the existing biases in its training environments.
By decoupling the gears of moral choice from human interaction, whether by choice or by inertia, humanity is being removed from the mechanisms that amplify moral and legal action (or, in some perverse cases, amplify the biases intentionally)
to build on your point, we only need to look at another type of entity that has a binary reward system and is inherently amoral: the corporation. Though it has many of the same rights as a human (in the US), the corporation itself is amoral, and we rely upon the humans within to retain moral compass, to their own detriment, which is a foolish endeavor.
even further, AI has only learned through what we've articulated and recorded, and so its inherent biases are only that of our recordings. I'm not sure how that sways the model, but I'm sure that it does.
And honestly, that's a ridiculous claim. Two very popular websites I can think of right away are Pinterest and VSCO. (Perhaps VSCO on the web isn't as popular as the mobile app, but the company continues to use masonry as the design evolves.)
Putting aside mental illness, what makes a person a criminal? What circumstances led to their decision?
Sure, protect the shift worker who has little power and therefore little blame. But, who's to blame for society's ills then? Calling for more arrests and incarceration has never, ever solved the underlying issues. What Garry Tan doesn't realize or won't admit is that he is part of the wealthy caste that caused this. Now that he got his money? Lock all of these bad people up!
Society's ills? I think we're talking about just SF. The rest of the nation seems to handle things at a better rate, per my comment about NYC where I live. So what is special about SF except lax rule of law?
I agree heavy sentences and private jails and, what is clearly a planned system to take offenders and make sure they can't re integrate into society are terrible and we need reform in the jail system.
Policing, as a deterrent, and arresting for petty crime however are effective. The threat of punishment is an effective deterrent. The fact we have a draconian system after that point is unfortunate, but if I understand SF doesn't even present the threat.
> The threat of punishment is an effective deterrent.
> [Police] arresting for petty crime however [is] effective
Incorrect. These are myths that people fall back on when they want to advocate for the same policy positions that have produced worse social outcomes for at least the past 40 years. The same thing, over and over again, driven by the belief that punishment must be involved to deter crime.
The threat of punishment has zero effect on deterrence. "Certainty of being caught" apparently does, but as you build a police force that makes getting caught of committing a crime a certainty, you have already fallen into the same trap America (and her cities) fall into over, and over, and over, and over, and over, and over, and over and over and over and over and over and over again.
"1. The certainty of being caught is a vastly more powerful
deterrent than the punishment.
Research shows clearly that the chance of being caught is a vastly more
effective deterrent than even draconian punishment. "
That's my point, and it was the first bullet point in what you linked. Can't be worried about being caught if there is no policing.
> "Certainty of being caught" apparently does [provide a deterrent effect], but as you build a police force that makes getting caught of committing a crime a certainty, you have already fallen into the same trap America (and her cities) fall into over, and over, and over, and over, and over, and over, and over and over and over and over and over and over again.
Police aren't a solution. They're a problem. That's not to say there's no UPSIDE. It's to say there are trade-offs and I think they have been proven by history to be unfavorable trade-offs in the long view.
