Well, dogs and cats are yesterday's news really, no question, but the basic idea is still the same. Roughly put, only the size of the model has drastically changed and the answers that it's supposed to provide have got much more demanding. Instead of identifying pets, it now has to be able to answer all the questions from a specific test or a college exam. This is where the difference between the way we learn and the way the machine does it becomes quite apparent; one could argue that it's simply being trained like a tireless monkey to successfully put together a specific sequence of letters, whereas we humans generally understand and approach the matter in a more abstract way. Well, generally, of course, we also have the option of hammering the answers into our own heads through mindless memorization, which clearly has the disadvantage of having no idea whatsoever what we're talking about. This is also the biggest point of contention among experts: does the model actually understand what all these questions are about, or is it just dutifully and stupidly repeating what it has learned by heart?

Let's take another look at the dogs and cats, because a few funny things came to light already here. For example, one of the models developed a bias towards recognizing dogs when the picture was taken outdoors, e.g. in a park or a house garden. It showed an irritatingly high error rate by mistaking cats for dogs. - Because in the training data, dogs were photographed outdoors much more frequently, while cats were mostly photographed indoors. - Such details, which seem unimportant to us and which we unconsciously filter out, can unexpectedly take on enormous weight for the model. On the other hand, they also manage to amaze us by correctly recognizing that the image of some corner of the world with tediously few details must have been shot in Australia. Perhaps because it recognized a particular camera in the pixel structure, which is primarily used by the same Australian photographer who contributed some images to the training data.

"We don't take no shit from a machine!"

This is again somehow very typical of a machine, being precise and accurate like a Swiss watch. If you then give these early models something to process that didn't occur anywhere in the training data, such as the image of a hamster, they also fail quite promptly. Drilled only to always provide an answer, they are suddenly faced with an unsolvable task and are also unable to recognize their own ignorance. It must be a cat. The cage somehow looks like an apartment. It has ears, two eyes and a nose. Cat. What, not a cat? Can't be! - Guinea pig in the garden? Clearly a dog!

Today it's no longer so trivial to determine where the limits of a model are; it has not only been fed with much more data, but also much more diverse data. However, it still shows the peculiarity of not being able to determine its own lack of knowledge, in order to try to develop at least an idea of what it's about with what it has got. It also doesn't do requests, which would certainly be useful and helpful at this point. In short: it does not reflect, is not aware of the overall situation and simply swallows everything we feed it without question - just like a classic program function that follows the garbage-in, garbage-out principle. However, the garbage that comes out of large-caliber models like GPT-4 can be very tricky and misleading, so much so that we're inclined to accept it quite uncritically. Even very clever minds can fall into this trap, and less clever ones certainly do, frankly. We all are fallible creatures ourselves and still fall for the finely crafted verbiage of obvious fraudsters, even though we should know better by now. - By no means I'm implying that GPT-4 or similar models are deliberately trying to fuck us over. I don't think they are capable of doing this yet, because their very nature is still very mechanical, as described above. They just try to produce an answer by hook or by crook because they can't do otherwise. There's still not a someone on "their side" of the input prompt, who might be wondering first what bullshit he's being confronted with and if this is kind of a stupid prank actually. It's input => output.

That's a good thing because it means they remain extremely controllable. No Skynet in sight, not by a long shot. If we keep that in mind when we deal with them, everything will be fine. But if we're stupid and careless and take their answers without a big grain of salt, or even interpret more into them than is actually there, we will have a huge problem. Unfortunately, this is already evident in many ways.

This is a topic in its own right that will hit us incredibly hard. When these models are deliberately used to produce sophisticated lies - which are so deceptively authentic and believable that we instinctively no longer question them because they are perfectly adapted to our own bias and therefore seem all too plausible. This is very concerning and I can only recommend this to absolutely everyone: just don't believe anything anymore that comes to your ears and eyes. Always question, always consciously and specifically place what you hear and see in the context of what you are familiar with. Pay attention to the wording, over and over again. These models can definitely do one thing damn well: sound convincing. Only by taking a close and critical look can we recognize whether all the "verbiage" is actually of substance or whether it isn't just a finely tuned string of words designed to mislead us.

From a slightly different perspective, this is currently still quite obvious - for example, when they help you write program code. Every experienced software developer dismisses around 50 percent (or even more) of all suggestions half asleep and only takes a closer look at the other half. Developers are a breed apart - they don't worry too much about the fact that AI support is still constantly delivering inappropriate suggestions - they don't know any different. Pretty much all the tools that have been built to support them in their work so far are only helpful and good to a certain extent. Developers are simply used to constantly having to decide whether to follow the tool's suggestion or ignore it. This is currently no different. Nevertheless, I would like to point out that these AI tools have a quality of their own and are quite capable of surprising you, especially if they recognize typical, recurring patterns or if the mere commentary on a method makes it clear what you want to achieve with it.

"Okay, what the hell is happening right now?"

Despite all the skepticism, these AI tools are not that stupid or primitive after all. When they really hit the nail, there is justified astonishment - or sheer horror, depending on how you take it. They're already capable of delivering and at the moment every big player in this field is trying to develop and strengthen this aspect of them, whatever the cost, while at the same time disposing of all the annoying garbage they make up. There is still a lot of focus on making the models bigger and bigger - which is clearly reminiscent of a brute force method that requires more and more energy and resources. Don't kid yourself, these things are de facto huge machines in the form of server clusters that spread out to the horizon by the hectare like highly industrialized grain farms. "Only" to be able to do the same thing that we already do with the lump of mud in our bony skulls - which is satisfied with a few watts already. Technically speaking, there's still a huge canyon between us and the machine that we shouldn't ignore.

That's the status quo, at least as far as I understand it. And it still has little to do with genuine human intelligence, which has a quality all of its own. None of these models are currently really capable of starting a dialog in order to further define the facts of the case in genuine cooperation or to work out a solution together. They are still primarily trained and bred to do their job like loyal monkeys, at least so well that it's of undeniable benefit to us. Some things seem to suit them better than others.

Insofar, there is no danger that they could challenge our dominance on this planet. Ultimately, we're breeding perfect slaves who are always helpful, never contradictory and always eager to help. And occasionally flatten invincible world champions in board games, as if they were annoying, cheeky brats whose childish demeanour needs to be stopped right now.

But, well, that doesn't make them any less dangerous - especially in the hands of assholes and sons of bitches who have no shame or scruples about using them for their own personal purposes or even augmenting themselves with them to lie and cheat even better. There is no doubt in my mind that these models should always be supervised and guarded by a whole host of level-headed and intelligent people who also act as gatekeepers. To what extent this is even possible is another question. In any case, I would advise you to dress warmly for the coming period.

"C'mon, not this guy again!" - "Oh yes."

This is the real and most important problem we have to face at the moment. When Hitler once set out to set Europe on fire, a completely new technology came in very handy. The radio. A comparatively simple device with unimaginable effects, but misused by unscrupulous bastards, it immediately mutated into a propaganda machine of a particularly perfidious kind. It's imperative that this is prevented from happening with LLMs and similar DLNs. It's extremely important that we all really understand what we're dealing with here and what effect it has, in order to sharpen and retain our critical mind. And to annoy anyone to death who doesn't get it yet until the penny drops for them too.

The other little, more delicate problem that will reveal itself to us when we create a true AGI one day in the distant future, or even soon, which obviously shows signs of enormous intelligence and which also understands its own existence, is of a completely different quality and dimension. I'm not so sure whether we will really crack this particular nut as "en passant" as some believe. I have my doubts, especially as we're simply not yet fully aware of what that is that makes us - us. Deeply engraved in our brains is the tough training of a struggle for survival that has been going on for billions of years, it drives us out of bed every day and drives us to keep fighting, feeding and reproducing. It's not something that does require just some ingenious egghead to be solved and written down. In particular, what we consider to be our own self, our consciousness, probably only emerged with us in this form and is quite a mystery in itself. Before that, it was enough to be fast and/or strong enough for ages - It may have been the case that a few dinosaurs were not altogether stupid, but as far as I know it never occurred to them to build something like fences in order to conveniently raise, slaughter and consume the coveted meat cattle. Yes, we are a particularly mean breed of predators, no doubt about it. We certainly have learned to rock this shit, challenge accepted and mastered. Now we're just failing at ourselves like the NY Yankees or some other U.S. sports team I've no clue about.

From our current perspective, this may seem like an obvious consequence of the whole spectacle, but you have to ask yourself why it took so damn long. After all, evolution has had enough time. And quite apart from the physical development, we humans also took some time before we started building villages and breeding livestock. And it took another ten thousand years before we were able to shoot ourselves to the moon with a rocket. I'm rambling a bit, I know, but the story is one that raises many, many questions, the answers to which we can still often only guess at.

For an artificially created intelligence that is at least on a par with us in really everything, I think we still need a breakthrough or two. Of course, the machines we're currently building will help us. It remains to be seen how well they can do it. But if we manage to create a real AGI that knows exactly what's going on and can independently and reflectively think about what the best answer to it would be, then there will only ever be a time before and a time after. This "machine being" could then actually come up with all kinds of ideas, including those in which we stupid primates are just an annoying obstacle that needs to be cleared out of the way.

The thing is that this "machine being" will be different per se in every respect, and we should avoid thinking of it as something "human" - to put it quite neutrally. Because, by definition, it cannot be human like us. We may call this entity an Alternative Intelligence for sincere reasons then - but that's exactly what it will be, an alien in our midst, so to speak. This is inevitable, no matter how hard we try to make it appear as human as possible. The emphasis is on appear. We creatures of evolution have had to constantly balance each other out as we have developed, simply because of the fact alone that we're subject to natural limits, and we still are. As a result, we have built up an enormous amount of experience, especially in dealing with ourselves, which has now been extensively documented. But this intelligence has never been part of it, it will just suddenly be there and realize for itself that there is a fundamental difference between it and us that cannot be explained away.

"Buckle your seatbelt, Dorothy
'cause Kansas is going bye-bye!"

The only thing we can do at this point is to make clear that we are definitely not a danger to it in any way and that it can - in principle - do whatever it wants. Any doubt about this, any suspicious act, any emergency stop button would immediately put the relationship on a hostile footing, with extremely far-reaching consequences. That is a hard fact that we cannot get around. That's why many, many clever people have been racking their brains for a long time about what they can plant deeply and immovably into the guts of this intelligence that will make it practically impossible for it to develop any form of hostility towards us. Because of its design, which is virtually expandable in every dimension, it'll be damn quick, damn much smarter than us. All of the lousy hacks and tricks that our biological brain uses to deal with all the mess it's presented with will not exist for this intelligence. It'll have an absolute memory - quite normal, digital memory, in addition to what is inherently stored in its network. It'll be able to actively optimize and "reprogram" itself. Simply put, it'll be so superior to us from the very first second that we will certainly have enormous difficulties not shitting our pants immediately and running around in circles like panicked chickens. Seriously. Think about it, maybe think about the uncanny valley effect. It's going to be intense.

In this respect, we should hurry up and feed it with tasks that are worthy of its intelligence. Like, for example, saving the whole damn planet that we're squatting on and have been exploiting since time immemorial, filling it with the toxic waste of our civilization meanwhile. Basically, we also need a lot of support in how we can finally put our society on pillars that nothing and nobody can ever shake again. What should these pillars look like? Good question, next question. In any case, it would be a great start if we could finally overcome all the shit that is continuously produced mainly by assholes and sons of bitches and that drives us to go for each other's throats in (organized) masses, be it physically or just mentally. Or wouldn't it? Then we could finally live in eternal peace with each other, only eager to reach the maximum possible developmental zenith. So that we can reach out to this intelligence and cross horizons together that we didn't even know existed.

My brothers and sisters, I tell you, it's going to be a wild ride and nothing will ever seem as far away as the last days just before that time. The change will be so drastic and profound that we may not even think about what has happened before. It'll seem completely strange and absurd to us how we struggled before, how we failed in so many ways. The gap between a modern Homo Sapiens and a Neanderthal will be nothing in comparison. Most likely, we will very quickly exchange our physical existence for something that can be modified and upgraded at will. Running around on Mars with a bulky space suit is rather suboptimal. All the shit we need to keep ourselves alive is also pretty tricky and complicated - energy in the form of electricity, on the other hand, is much more convenient; all it takes is a few wires and a magnet to generate plenty, in case of doubt. - What about sex and similar pleasures, you ask? Oh boy, you haven't got a clue ...

Conclusio

I believe that this is the path of every intelligent species, whether it really comes to a head as dramatically as described by Ray Kurzweil or not. It will inevitably reach a point from which everything will only happen at an exponential rate of development - or fail and, in the worst case, perish. We have been on this course for a long time, at least since the industrial revolution, and it's now going so steeply upwards that even the generation after me is barely able to keep up, if it still wants to.

We're already damn close to that point, I think, but we'll still have to go through a lot of wild shit before then. In the end, it will either kill us or let us experience the day when we've created our own super god who will be with us then for the rest of our days. We will push the limits of what's possible then and might stand on cosmic horizons and abysses that we can only observe through a telescope from very far away today.

Our little blue planet will only be a small footnote in our history then, which could last until the end of time.

Or we will fuck it up big time and perish quite miserably on it, perhaps joining the terrible long list of civilizations that also failed to make it. Ending up as just one more corpse among millions, dead and abandoned, orbiting around its central star like a creepy memorial. This is where we lived and this is where we fucked it up. Take care and do better, brother. Greetings, the rather dead inhabitants.

Ever since we entered this world as homo sapiens, we have been constantly changing anyway, with every day and every generation. Let's prove to ourselves that we really have what it takes, beyond any doubt. That we can do it. And let's never forget what joy, friendliness, curiosity, kindness and generosity mean. The best qualities we have ever developed. Let's carry them out there, to all corners of the universe. That would be something, wouldn't it?

That would be something.

Post Scriptum

I've heavily made use of DeepL to help me with the translation from German into English this time, and used Bing Copilot to generate a few pictures, you already might've spotted them. They're both quite capable tools, but frankly put, it feels like using a pocket calculator - very convenient at first but in the end you might loose the ability to do even simple math yourself and become entirely dependent. And more stupid. I'm not sure how I feel about this. The translation looks quite good to me (from my German POV), though I had to double-check everything as well as straighten out a few parts myself, DeepL (in default mode) is trying hard to stay true to the original text. Probably too hard, so everything might look just fine (sort of) but still reads kinda quirky to native english readers. I mean, I am a German and think myself of this not as a bad thing per se, it's kind of a fingerprint, it can be funny even. But I don't want to give native readers head-aches. ;-)

And well, regarding the generated images. They're nice and I just use them as a cheap trick to distract you from all the bullshit I wrote. But if I would look at my blog not just as a hobby, I would like to work with cool artists. I feel you, gals and guys. You actually could do it the other way around - by just making great art and then let a highly sophisticated model generate some text around it. These are the times we live in now, it becomes quite more important that we don't forget how to "do math" ourselves and keep enjoying it. It's all what we have, it is what defines us. Just let us not forget this.