AIs gunning for our precious freelancers

[zompist]

Regarding this specific problem, it's totally possible to infer statistically likely details from contextual information.

"Statistically likely", yes. If you flip ten coins, it's statistically likely that they're half heads. What they actually are can only be determined by actually flipping them.

But this is not the core of the issue. Arguments like this are motivated by the theological instinct that complexity cannot arise unless it was created and information cannot be obtained unless it was given.

No, arguments like these are motivated by being interested in what is true. You don't even seem to understand the problem, or what your machines are actually doing.

[rotting bones]

So weather forecasting is impossible now? How is that different from guessing the folds that led to a set of pixels?

---

A neural network is a technique that lets you approximate the constants in a system of equations that model your training data.

The details differ from system to system, and several interpretations are possible. Linear equations with nonlinear activations are often interpreted as a region in n-dimensional space bounded by hyperplanes. Convolutions are usually used on images to aggregate neighboring information. Transformers specifically have been interpreted as interacting systems of particles: https://arxiv.org/abs/2312.10794 Etc.

I don't understand why non-specialists think they have special insight into this field.

[zompist]

So weather forecasting is impossible now?

Are you unable to read? Is it too hard to respond to the things I actually say, so you make up shit instead?

How is that different from guessing the folds that led to a set of pixels?

I gave you several examples where statistical methods don't work. You ignore the facts you don't like and just pretend everything is weather reports. That's not rationalism, it's rationalization.

[rotting bones]

I was responding to the last critical post:

i'm pretty sure one can't get more data than one started with… by "enhcancing" even if you increase the amount of information content.

Regarding your examples, it's not that statistical models "don't work". All models have biases that have to be gradually corrected. Even the standard model of physics is understood to be probably wrong in an absolute sense. On the other hand, newer models wouldn't turn Obama white if that is how the model was trained.

What you should be arguing is that before rushing to use AI models in practical applications, democratic processes should be implemented to ensure that society is comfortable living with the consequences given the AI model's performance. For example, an AI model's prediction cannot be used in a situation that requires the verdict to be "beyond reasonable doubt". There is always some doubt in any statistical technique. On the other hand, AI models are good at being image filters for artistic purposes, rewriting text to be in specific formats, etc.

I don't understand why you keep accusing me of not reading your post. I listed this use case under "amazing applications". Under useful applications, I mentioned cancer diagnosis. Doctors admit that those models are better at identifying malignant tumors than they are.

[zompist]

I don't understand why you keep accusing me of not reading your post.

Because you skip over the parts that refute your claims and make up shit I didn't say. Case in point:

All models have biases that have to be gradually corrected. [...] On the other hand, newer models wouldn't turn Obama white if that is how the model was trained.

I explicitly addressed this in my original post; I said it wasn't a matter of adding more Obama to the training data. You skipped over the clear examples I provided of filling in detail wrong, and you missed the point of the Obama example exactly as I warned not to.

On the other hand, AI models are good at image filters for artistic purposes [...]

Sure, and what does that have to do with your claim that hallucinated data is real? Again, I said AIs have uses in my post. What magic process do I need to have a statistical chance of you reading my words?

What AI can do, of course, is add plausible details that don't matter. That can be common enough-- e.g. upscaling images for a video game.

For example, an AI model's prediction cannot be used in a situation that requires the verdict to be "beyond reasonable doubt".

Not putting people in jail because of hallucinated data would be a start, yes. There is an immense middle ground of cases where the AI is not putting someone in prison, but it's creating false data where we expect real data. E.g. should image processors be used to enhance historical photos? If it's for a stage backdrop or your conworld or desktop wallpaper or something, sure, go ahead. If it's for a museum that is supposed to depict what is actually there, no, because you do not have a Star Trek enhance button. You have a tool that makes photorealistic fakes based on large input sets. If all we have as actual data is a small photo, then too bad, all we have is a small photo.

[Travis B.]

The matter here is that while on one hand AI can't add information that isn't there -- that's physically impossible -- but AI can analyze data in such a fashion to perceive things that normal humans cannot -- like the example of cancer diagnoses. In the former case, things like AI turning Obama White are an effect of AI not being able to do anything beyond what it is trained for and with, and if you images of US presidents are all White, garbage in garbage out. In the latter case, though, if you train your AI on extensive training sets including data tagged as non-cancerous and data tagged as cancerous, you can potentially get an AI that can diagnose cancer better than a human can, as your training sets can be far larger than the number of patients an oncologist might see in a career.

[Raphael]

Found this on Bluesky:

Of Nerniacular Latin to an Evoolitun on Nance Langusages!

https://bsky.app/profile/bretdevereaux. ... uxswwrzk2e

Any ZBB members interested in constructing Nerniacular Latin and the Nance Langusages?

[malloc]

It always baffles me how people using AI images never seem to check whether they make sense. So often I see thumbnails on youtube videos and such with utter nonsense like dinosaur-mammoth hybrids with five legs and two mouths. It would only take one minute to check whether the image looks good and generate a new one if necessary.

All mockery aside, I feel like we don't appreciate how amazingly powerful computers have become over the past few decades. They say even the greatest chess player could never defeat an average chess computer, for instance, and no human could match a calculator in performing arithmetic. I feel like we achieved artificial intelligence and perhaps even superintelligence ages ago. If you told someone in the fifties that computers would one day steamroll the best chess players, they would surely consider that superintelligent AI at least.

All that differs of course from artificial general intelligence that could replicate any human cognitive ability. Yet it seems increasingly difficult to find any such abilities that computers cannot replicate these days. Combine everything from generative AI to calculators to chess engines and everything else computers can handle now into one machine and the result would come remarkably close to general intelligence.

[Raphael]

All mockery aside, I feel like we don't appreciate how amazingly powerful computers have become over the past few decades. They say even the greatest chess player could never defeat an average chess computer, for instance, and no human could match a calculator in performing arithmetic.

Long before computers in the modern sense existed, cars and trains were able to drive faster than the fasted human runners.

I feel like we achieved artificial intelligence and perhaps even superintelligence ages ago.

Feeling something doesn't make it true.

If you told someone in the fifties that computers would one day steamroll the best chess players, they would surely consider that superintelligent AI at least.

If so, then only because people back then didn't really think through what "intelligence" means.

All that differs of course from artificial general intelligence that could replicate any human cognitive ability. Yet it seems increasingly difficult to find any such abilities that computers cannot replicate these days. Combine everything from generative AI to calculators to chess engines and everything else computers can handle now into one machine and the result would come remarkably close to general intelligence.

No, because it would still lack basic common sense. It might be able to calculate all kinds of things, but it wouldn't be able to decide on its own what it wants to calculate in the first place.

[Travis B.]

I would say that the LLM's and like we have today aren't "intelligent" not because intelligence is some special attribute of human brains or at least biologically-based neural systems (there is no reason why it should be) but because they are essentially one-way large-scale pattern-matching systems -- they "learn" based off being trained on absolutely massive datasets, but essentially lack the ability to complete the loop -- they do not learn from observing the results of their own actions and build upon them, and they do not have true introspection.

The thing is, there is nothing stopping us from going the next step and building these things into future AI's -- however, the kind of people who put massive amounts of funding into the development of AI are unlikely to do so for the reason that they want mindless replacements for human workers and like, and seeking true AGI will mean creating something that is not mindless.

[Raphael]

I would say that the LLM's and like we have today aren't "intelligent" not because intelligence is some special attribute of human brains or at least biologically-based neural systems (there is no reason why it should be) but because they are essentially one-way large-scale pattern-matching systems -- they "learn" based off being trained on absolutely massive datasets, but essentially lack the ability to complete the loop -- they do not learn from observing the results of their own actions and build upon them, and they do not have true introspection.

The thing is, there is nothing stopping us from going the next step and building these things into future AI's

You mean, aside from the fact that currently, no one has any idea how to start doing that?

[Travis B.]

I would say that the LLM's and like we have today aren't "intelligent" not because intelligence is some special attribute of human brains or at least biologically-based neural systems (there is no reason why it should be) but because they are essentially one-way large-scale pattern-matching systems -- they "learn" based off being trained on absolutely massive datasets, but essentially lack the ability to complete the loop -- they do not learn from observing the results of their own actions and build upon them, and they do not have true introspection.

The thing is, there is nothing stopping us from going the next step and building these things into future AI's

You mean, aside from the fact that currently, no one has any idea how to start doing that?

If you asked me ten years ago whether I thought that in ten years' time we'd've demolished the Turing test, or even created an AI that could soundly defeat a top Go player (which happened in 2016) I would have definitely said no. Yet here we are. Given that, I am not going to place my bets against AGI. AGI may be a ways off, it may not be, but I don't think we collectively can easily judge this at this point.

Edit: corrected a mistype.

[Raphael]

I would say that the LLM's and like we have today aren't "intelligent" not because intelligence is some special attribute of human brains or at least biologically-based neural systems (there is no reason why it should be) but because they are essentially one-way large-scale pattern-matching systems -- they "learn" based off being trained on absolutely massive datasets, but essentially lack the ability to complete the loop -- they do not learn from observing the results of their own actions and build upon them, and they do not have true introspection.

The thing is, there is nothing stopping us from going the next step and building these things into future AI's

You mean, aside from the fact that currently, no one has any idea how to start doing that?

If you asked me ten years ago whether I thought that in ten years' time we'd've demolished the Turing test, or even created an AI that could soundly defeat the a top Go player (which happened in 2016) I would have definitely said no. Yet here we are. Given that, I am not going to place my bets against AGI. AGI may be a ways off, it may not be, but I don't think we collectively can easily judge this at this point.

I have no problem with saying "that might happen". My problem is with your assertion that "there is nothing stopping us from" doing that.

[Travis B.]

You mean, aside from the fact that currently, no one has any idea how to start doing that?

If you asked me ten years ago whether I thought that in ten years' time we'd've demolished the Turing test, or even created an AI that could soundly defeat the a top Go player (which happened in 2016) I would have definitely said no. Yet here we are. Given that, I am not going to place my bets against AGI. AGI may be a ways off, it may not be, but I don't think we collectively can easily judge this at this point.

I have no problem with saying "that might happen". My problem is with your assertion that "there is nothing stopping us from" doing that.

You seem to imply that you think that it is outside the range of our current capabilities. I am not so sure, and am unwilling to commit to such an assertion. (By what I said I meant that there is no fundamental barrier preventing AI developers from going in that direction.)

[Travis B.]

The next step, IMO, is defeating Steve Wozniak's Coffee Test -- that an AI in a robot can enter an unfamiliar house, find the kitchen, locate where the coffee is stored, find the coffee machine, figure out how to operate the coffee machine, and brew a cup (or pot) of coffee. Until an AI can do this, we can't say we have AGI, and there is no known AI which has beaten this test so far.

[Raphael]

The next step, IMO, is defeating Steve Wozniak's Coffee Test -- that an AI in a robot can enter an unfamiliar house, find the kitchen, locate where the coffee is stored, find the coffee machine, figure out how to operate the coffee machine, and brew a cup (or pot) of coffee. Until an AI can do this, we can't say we have AGI, and there is no known AI which has beaten this test so far.

Thank you for telling me about that. It's a great concept. And it's what I mean by "We don't know where to start". Does anyone have any kind of "plan of attack" for approaching this problem?

[Travis B.]

The next step, IMO, is defeating Steve Wozniak's Coffee Test -- that an AI in a robot can enter an unfamiliar house, find the kitchen, locate where the coffee is stored, find the coffee machine, figure out how to operate the coffee machine, and brew a cup (or pot) of coffee. Until an AI can do this, we can't say we have AGI, and there is no known AI which has beaten this test so far.

Thank you for telling me about that. It's a great concept. And it's what I mean by "We don't know where to start". Does anyone have any kind of "plan of attack" for approaching this problem?

To me, the big problem here is giving AI's a degree of "common sense" -- the kind that is acquired by growing up in and living in houses with kitchens and coffee machines, which is not likely to be easily acquired by simply training an LLM against human-language knowledge scraped from the Internet -- combined with programming an AI with spatial awareness. This to me is not insurmountable, it is not simply out of range of what we can do, it just requires finding a different way of training AI's against different kinds of knowledge than AI's are typically trained against today combined with designing AI's meant to autonomously operate in Real Life as opposed to in contrived environments and situations. The matter, though, is that the training data will not be easy to come by, because it will require acquiring data from operating robots in actual houses rather than mere web-scraping.

[malloc]

Thank you for telling me about that. It's a great concept. And it's what I mean by "We don't know where to start". Does anyone have any kind of "plan of attack" for approaching this problem?

All too easy. The robot needs to know how to walk, how to identify rooms by recognizing walls and doors and so forth, how to identify the kitchen (presumably by recognizing typical appliances), and how to identify the coffeemaker. We already have ways of teaching computers to recognize things visually and robots that can walk and manipulate objects. We would simply need to put all the pieces together.

[Raphael]

Thank you for telling me about that. It's a great concept. And it's what I mean by "We don't know where to start". Does anyone have any kind of "plan of attack" for approaching this problem?

All too easy. The robot needs to know how to walk, how to identify rooms by recognizing walls and doors and so forth, how to identify the kitchen (presumably by recognizing typical appliances), and how to identify the coffeemaker. We already have ways of teaching computers to recognize things visually and robots that can walk and manipulate objects.

You might be falling into the fallacy where people thing "This robot can do X. People who can do X can usually do Y, too. Therefore, this robot should be able to do Y!" But that's a fallacy, because robots aren't humans, and their skills don't work the same way as human skills.

And, you underestimate how difficult generalizing has traditionally been for robots and computers. For instance, if a robot can walk through a specific parcour, that doesn't mean it can work everywhere, under real-world conditions.

[Travis B.]

Thank you for telling me about that. It's a great concept. And it's what I mean by "We don't know where to start". Does anyone have any kind of "plan of attack" for approaching this problem?

All too easy. The robot needs to know how to walk, how to identify rooms by recognizing walls and doors and so forth, how to identify the kitchen (presumably by recognizing typical appliances), and how to identify the coffeemaker. We already have ways of teaching computers to recognize things visually and robots that can walk and manipulate objects. We would simply need to put all the pieces together.

The real problem, though, is that the robot will need to be able to operate in a wide range of houses with different layouts and kitchens with different layouts, be able to find coffee (and if the criterion is to make a cup of coffee, a coffee cup) that may be stored in any number of places in that house (typically, but not necessarily, the kitchen or pantry), be able to operate a range of different coffee machines with different designs, be able to dispense ground coffee into the coffee machine (and if the coffee machine is a filter coffee machine, find an unused coffee filter and fit that in its holder for the machine), if the coffee isn't preground recognizing that it isn't and locating a coffee grinder and operating it to grind the coffee beans prior to putting them in the machine, and if the goal is to pour a cup of coffee, pour coffee into the cup when everything is all done. This is much harder than it sounds. It requires a different level of knowledge and functionality on the part of the AI than any of our present-day AI's have.