Homophily, or the answer as to why there are AIs and LLMs

I don’t know if you’ve noticed, but it seems that when it comes to LLMs (popularly called AI models, though only the “artificial” part of the name has any application to the reality of things), depending on who you ask they are either the second-best invention of humanity (after fire, I presume, or maybe sliced bread?), or a scam ran by the the Tech Bros from Silicon Valley. There’s nothing in between, and indeed, how could there be? We are all living inside what I like to call: an economy of bubbles. Plural, as at any given time there are at least several different types of bubbles operating, be it social bubbles (these can be counted in hundreds), or economic bubbles, or any other bubbles you can think of. I suspect that the number of bubbles may be the result of an attention-based economy where every entity is competing with millions of others to steal your attention for at least a second, so that you click on the thumbnail/title/image (choose what you will). And I here also stand proud in the line of thousands of other articles on LLMs to try and grab your attention for long enough so that you read what I have to say on the subject.

And to be honest, there are many things I have to say on the topic of LLMs, most of them unflattering. To put it bluntly, in my humble opinion, LLMs are a total waste of everything (order is random and the list is meant as an example and not to be exhaustive):

• Time: because I wanted to learn something today, but instead I caught a glimpse of an article about how LLMs are so powerful that we all would have perished already if it weren’t for the heroic AI companies warning us about the RISKS, and it got my blood boiling.

• Attention: as I said, I had other plans for today’s evening, yet here I am furiously typing away to get my frustration out of my chest, or else I go up in flames; and this one is especially painful when you realise that we live our lives through attention, so spending it on something so utterly useless like LLMs seems awfully irresponsible.

• Earth: which by all accounts is the only thing in this universe that can actually sustain us, but of course, in this twisted turn of fate we are letting hyper-scalers destroy it to try and build even more data centers (this will not happen, luckily VC’s pockets are not that deep, but the environment will still be destroyed, even if all of these investments fail);

Let’s stop with these three examples. You know where I stand now.

Of course, we could say that everyone is entitled to their own opinion, and that is true, though most people probably don’t care deeply enough about the world they live in to even form an opinion that could be called their own. For better or for worse, people seem to be repeating statements made by others without much reflection. And because nothing keeps us going as much as an adversary (real or imagined) within all that division, the point is inevitably lost.

LLMs are a perfect example to illustrate this issue. There are scientists who actually believe that LLMs can be the gateway to AGI (Artificial General Intelligence). There are scientists who believe that’s not the case. As always with science, there’s much faith to be had (probably much more than rationalists would like to admit). Then there are people who listen to the scientists from one camp or the other and take their opinion as a proof of… well, of something, nobody knows what is there to prove, because nobody can agree on what intelligence is.

For me, personally, you need to be a living creature to have intelligence, so no matter how advanced a computer program is, or how well it can impersonate my dead grandma (well, that’s not a good example, because she never posted anything online, so there are no cyber breadcrumbs to regurgitate and extrapolate from, but the image of a lovely old lady does its job here) I will never concede that it’s intelligent. And yes, that is a matter of belief, as many things in our modern world are.

Then we have all the tech people working inside AI companies (I apologise in advance if you are one of them, please don’t be mad at me: I have a close relative working in a company like that and we still managed to talk about LLMs, though admittedly some amount of alcohol was involved and we agreed to disagree in the end) who generally think themselves to be the experts on LLMs - which they are - unfortunately, in most cases only on LLMs - which is not enough, I am afraid. Indeed, they are the group that I would like to write about today.

On paper, people working on LLMs, with these algorithms, training them, testing them, etc., should be the ones you look up to for answers and guidance, shouldn’t they? They have the knowledge and skills necessary to judge whether a model like Mythos is really as dangerous as Anthropic says it is.

And let’s put aside the fact that not everyone needs to be this great, selfless person who always does what’s best for humanity. Let’s not even go there, because we would also have to point out that there are real financial incentives if you convince CEOs that your model will be able to halve the number of people they employ (you know how it is: human = cost, and cost doesn’t look great in the financial report). For the sake of this short piece, let’s even ignore the fact that LLMs do not yield anything. Really, if you need to spend 200$ on a subsidised subscription that at some point will cost you a 1000$ (or more), it would make more sense to invest in tomatoes. You will still need to use soil, water and energy, but on incomparably smaller scale, and if you play your cards right in a few weeks, you will be able to eat a tomato soup. That’s what I call a return on investment. And the friends you make on your way to the tomato soup? Priceless.

So let’s put all of that aside and just look at one of the many problems connected with or caused by LLMs: the sheer size of division in opinions when it comes to what LLMs are capable of.

The tech guys will always say that the normies simply don’t understand. They cannot imagine the possibilities (by the way, in this example, I am one of the normies). That if the normies knew what they - the experts - know, or could envisage how brilliant these models must be because they are already oh so very impressive, and we don’t even have access to the state-of-the-art models, if the normies knew, their minds would be blown away (either figuratively or in reality, because we all know that Mythos is coming for us).

Then, when this tech person meets a sceptical normie (indeed, the worst kind of human), they often come out on the other end of such a meeting deeply frustrated that their warnings are downplayed. They would blame the normie and think to themselves that they have the duty to go online and educate the rest of the population, so that maybe at least the legislators will understand how big and important the LLMs like Mythos are.

For quite some time, I couldn’t understand why the tech people who will go out of their way to lecture normies about their lack of understanding, how on earth do they not see their own biases?

They would argue with the world that every time an LLM model beats some sort of benchmark that was supposed to prove that it is intelligent, the benchmark is changed to something else. They will talk about the hypocrisy of normies who constantly move the goal post. They will burn with the brightest fire only a zealot's heart can produce, and argue to the end of time that we are on the cusp of greatness with these state-of-the-art models. All of that, while ignoring that the LLMs were trained on the information that is checked for in the said benchmark, and that the companies that built these models also test them on the benchmarks, and the results tend to be less impressive if an independent lab tests the same models.

And then I found this beautiful term - homophily - and everything became clear:

“Homophily is a concept in sociology describing the tendency of individuals to associate and bond with similar others, as in the proverb ‘birds of a feather flock together.’ The presence of homophily has been discovered in a vast array of network studies: over 100 studies have observed homophily in some form or another, and they establish that similarity is associated with connection. The categories on which homophily occurs include age, gender, class, and organisational role. […] (Wikipedia)”

Apparently, specialists are not immune to homophily. What is the result then? Well, to put it simply, it means that you lose the ability to see the forest for the trees. Or more directly: because you mingle with people who share your background, convictions, and degree, you also share their point of view on different subjects, as well as errors in their logic.

That’s why I always have the feeling that I am losing my mind when I try to follow the LLMs debate (whichever facet of it happens to be currently trendy). For if you only listen to the tech guys, you will come to the conclusion that not only are LLMs AI, but also that they are on the cusp of becoming AGI. And once that happens, we will all lose our jobs. But worry not, for that will be the gateway to the Garden of Eden. In there, the all-powerful LLMs and/or ‘AGI Tree of Knowledge’ will grow right in the centre, very much reminiscent of the Yggdrasiltree.

And what’s next? - you may ask. Well, the Tech Bros are not too sure about that, but I guess it’s safe to assume that we - the peasants - will have to perform some sort of ritual to be able to partake in the riches that the few of the chosen ones may be willing to share with us.

The problem I have with this narrative is the following: if we skip the context or the ingredients that make up this whole endeavour, we end up only with a distorted image.

An image that is very similar to the one an LLM might create when you prompt it to extrapolate an image of a woman reading a book in the ruins of a temple under the full moon, surrounded by the creatures of the night (as in the case of the image I chose for this article - do try to find all the things that are wrong with it).

Or worse still, you will then ask it to change it to a short movie.

Truly, the context inevitably gets lost if we put aside all the things that make up life.