LUIs, “Approved AI” and controlled speech

If we’re not careful here, this whole AI "revolution" could become the “Great Homogenisation”

I’ve been travelling a lot recently, so this essay is a little belated, but still relevant because we need to separate real risks from fake risks when it comes to “AI.”

Yes. The world is changing before our very eyes. Generative AI is a paradigm shifting technological breakthrough - but probably not for the reasons you might think, or have been told by the mainstream.

You’ve probably heard something along the lines of “AGI is around the corner” or “now that language is solved, the next step is conscious AI”.

Well…I’m here to tell you that those concepts are both Red Herrings. They are either the naive delusions of nihilistic technologists who believe god is in the circuits, or the deliberate incitement of fear and hysteria by more malevolent people with ulterior motives.

I DO NOT think AGI is a threat or that we have an “AI safety problem” or that we’re around the corner from some singularity with machines. 

BUT…

I do believe this technological paradigm shift poses a significant threat to humanity - which is in fact, about the ONLY thing I can somewhat agree on with the mainstream - but for completely different reasons. 

To learn what they are, let’s first try to understand what's really happening here.

Introducing…the Stochastic Parrot!

Technology is an amplifier. It makes the good better, and the bad worse.

Whether that technology is a hammer, used to build a house or beat someone over the head with, computers can be used to document ideas that improve the world, or they can be used to operate CDBCs that enslave you to crazy communist cat lady’s working at the ECB.

The same goes for AI. It is a tool. It is a technology.

It is NOT a new life-form, despite what the lonely nerds like Yudkowski so desperately want to believe.

What makes Generative AI so interesting is not that it is sentient, but that it’s the first time in human history that we are “speaking” or communicating with something other than a human being, in a coherent fashion. The closest we’ve been to that before this point has been with…Parrots.

Yes - a PARROT!

You can train a parrot to kind of talk, and talk back, and you can kind of understand it, but because we know it’s not really a human and doesn’t really understand anything, we’re not so impressed.

But Gen Ai…well that’s a different story. We’ve been acquainted with it for 6 months now (in the mainstream) and we have no real idea how it works under the hood. We type some words, and it responds like that annoying, politically correct, midwit nerd you know from class…or your average Netflix character.

In fact, you’ve probably even spoken with someone like this during support calls to Booking.com, or any other service in which you’ve needed to dial in, or have a web-chat with.

So you’re immediately shocked. “Holy shit. This thing speaks like a real person” you tell yourself.

The English is immaculate. No spelling mistakes. Sentences makes sense. It is not only grammatically accurate, but semantically too.

“HOLY SHIT! It must be alive !!!!”

Little do you realise you are speaking to a highly sophisticated stochastic parrot.

It turns out language is a little more rules based that what we all thought, and probability engines can actually do an excellent job of emulating intelligence through the frame or conduit of language.

The law of large numbers strikes again, and math achieves another victory!

But…what does this mean?
What the hell is my point?

That this is not useful? That it’s proof it’s not a path to AGI?
Not necessarily. On both counts.

There is lots of utility in such a tool. In fact, the greatest utility probably lies in its application as “MOT”, or “Midwit Obsolescence Technology”. Woke Journalists and the countless “content creators” who have for years been talking a lot but saying nothing, are now like Dinosaurs watching the comet incinerate everything around them. It’s a beautiful thing. Life wins again.

Of course, it’s also great for ideating, not bad with summarising (sorta kinda), coding faster, doing some high level learning, etc.

And from an AGI and consciousness standpoint, WHO KNOWS! There mayyyyyyyyyyy be a pathway there, but I’m not holding my breath. I think consciousness is so much more complex, and to think we’ve conjured it up with probability machines is some strange blend of ignorant, arrogant, naive and…well…empty.

So what the hell is my problem and what’s the risk?

Enter the age of the LUI

Remember what I said about tools.
Computers are arguably the most powerful tool mankind has built.

And computers have gone through (roughly) the following evolution:

1. Punch Cards

2. Command Line

3. GUI, ie; Point and Click

4. Mobile, ie; Thumbs and tapping

And now, we’re moving into the age of the LUI, or “Language User Interface”

This is the big paradigm shift. It’s not AGI, but LUI. Apps we interact with moving forward will likely have a conversational interface, and we will no longer be limited by the bandwidth of how fast our fingers can tap on keys or screens.

Speaking “Language” is orders of magnitude faster than typing and tapping. Thinking is probably another level higher, but I’m not putting any electrodes into my head anytime soon. In fact, LUIs significantly diminish the need for Nueralink type tech because the risks associated with implanting chips into your brain will outweigh any marginal benefit over just speaking.

In any case. This decade we will go from tapping on graphical user interfaces, to talking to our apps.

And therein lies the danger.

In the same way Google today determines what we see in searches, twitter, Facebook, Tik Tok and Instagram all “feed us” through their feeds; generative AI will tomorrow determine the answers to every question we have.

The screen not only becomes the lens through which you ingest everything about the world. The screen becomes your model of the world.

Mark Bisone wrote a fantastic article about this recently, which I urge you to read:

“The problem of “screens” is actually a very old one. In many ways it goes back to Plato’s cave, and perhaps is so deeply embedded in the human condition that it precedes written languages. That’s because when we talk about a screen, we’re really talking about the transmission of an illusory model in an editorialized form.

The trick works like this: You are presented with the image of a thing (and these days, with the sound of it), which its presenter either explicitly tells you or strongly implies is a window to the Real. The shadow and the form are the same, in other words, and the former is be trusted as much as any fragment of reality that you can directly observe with your sensory organs.”

Screen Grab

And for those thinking that “this won’t happen for a while”, well here’s the bumbling fools at the G7 making a good attempt at it:

And here are the OpenAI guys talking about “SuperIntelligence Governance” in their latest article, which of course comes with the usual dose un-defined words like “Safety,” “Responsibility” and “Compliance.”

The “Great Homogenisation”

Imagine every question you ask, every image you request, every video you conjure up, every bit of data you seek, being returned in such a way that is deemed “safe,” “responsible” or “acceptable” by some faceless “safety police.”

Imagine every bit of information you consume has been transformed into some Luke-warm middle version of the truth, that every opinion you ask for is not really an opinion or a viewpoint, but some in-offensive, apologetic response that doesn’t actually tell you anything (this is the benign, annoying version) or worse, is some ideology wrapped in a response so that everything you know becomes some variation of what the manufacturers of said “Safe AI” want you to think and know.

Imagine you had modern Disney characters, like those clowns from “The Eternals” movie as your ever-present intellectual assistant. It would make you “dumb squared”.

The UnCommunist Manifesto outlined the Utopian communist dream as the grand homogenisation of man. If only everyone were a series of numbers on a spreadsheet, or automatons with the same opinion, it would be SO much easier to have paradise on earth. You could ration out just enough for everyone, and then we’d be all equally miserable proletariats.

This is like Orwell’s thought police crossed with Inception, because every question you had would be perfectly captured and monitored, and every response from the Ai could incept an ideology in your mind. In fact, when you think about it, that’s what information does. It plants seeds in your mind.

This is why you need a diverse set of ideas in the minds of men! You want a flourishing rainforest in your mind, not some mono-crop field of wheat, with deteriorated soil, that is susceptible to weather and insects, and completely dependent on Monsanto (or Open Ai or Pfizer) for its survival. You want your mind to flourish and for that you need idea-versity.

This was the promise of the internet. A place where anyone can say anything. The internet has been a force for good, but it is under attack. Whether that’s been the de-anonymization of social profiles like Twitter and Facebook, and the creeping KYC across all sorts of online platforms, through to the algorithmic vomit that is spewed forth from the platforms themselves. We tasted that in ALL its glory from 2020. And it seems to be only getting worse.

The push by WEF-like organizations to institute KYC for online identities, and tie it to a CBDC and your Iris is one alternative, but it’s a bit overt and explicit. After the pushback on medical experimentation of late, such a move may be harder to pull off. An easier move could be to allow LUIs to take over (as they will because they’re a superior UX) and in the meantime create an “Ai Safety council” that will institute “safety” filters on all major LLMs.

Don’t think this won’t happen.

Today the web is still made up of web pages, and if you’re curious enough, you can find the deep, dark corners and crevices of dissidence. You can still surf the web. Mostly. But when everything becomes accessible only through these models, you’re not surfing anything anymore. You’re simply being given a synthesis of a response that has been run through all the necessary filters and censors. 

There will probably be a sprinkle of truth somewhere in there, but it will be wrapped up in so much “safety” that 99.9% of people won’t hear or know of it. The truth will become that which the model says it is.

I’m not sure what happens to much of the internet when discoverability of information fundamentally transforms. I can imagine that as most applications transition to some form of language interface, it’s going to be very hard to find things that the “portal” you’re using doesn’t deem safe or approved.

One could of course make the argument that in the same way you need the tenacity and curiosity to find the dissident crevices on the web, you’ll need to learn to prompt and hack your way into better answers on these platforms. 

And that may be true, but it seems to me that for each time you find something “unsafe” the route shall be patched or blocked.

You could then argue that “this could backfire on them, by diminishing the utility of the tool.” 

And once again, I would probably agree. In a free market, such stupidity would make way for better tools.

But of course, the free market is becoming a thing of the past. What we are seeing with these hysterical attempts to push for “safety” is that they are either knowingly or unknowingly paving the way for squashing possible alternatives.

In creating “safety” committees that “regulate” these platforms (read; regulate speech), new models that are not run through such “safety or toxicity filters” will not be available for consumer usage, or they may be made illegal, or hard to discover. How many people still use TOR? Or Duck Duck Go?

And if you think this isn’t happening, here’s some information on the current toxicity filters that most LLMs already plug into. It’s only a matter of time before such filters become like KYC mandates on financial applications. A new compliance appendage, strapped onto language models like tits on a bull.

Whatever the counter-argument to this homogenisation attempt, both actually support my point that we need to build alternatives, and we need to begin that process now.

For those who still tend to believe that AGI is around the corner and that LLMs are a significant step in that direction, by all means, you’re free to believe what you want, but that doesn’t negate the point of this essay.

If language is the new “screen” and all the language we see or hear must be run through approved filters,the information we consume, the way we learn, the very thoughts we have, will all be narrowed into a very small Overton Window.

I think that’s a significant enough risk to warrant some vigilance on our part.

We’ve become dumb enough with social media algorithms serving us what the platforms think we should know. And when they wanted to turn on the hysteria, it was easy. Language user interfaces are Social Media x 100.

Imagine what they can do with that, the next time a so-called “crisis” hits?
It won’t be pretty.

The marketplace of ideas is necessary to a healthy and functional society.
That’s what I want.

Their narrowing of thought won’t work long term, because it’s anti-life. In the end it will fail, just like every other attempt to bottle up truth and ignore it. But each attempt comes with unnecessary damage, pain, loss and catastrophe. That’s what I am trying to avoid and help ring the bell for.

What to do about all this?

Like I said earlier, if we’re not proactive here in, this whole AI revolution could become the “Great Homogenisation.” To avoid that, we have to do TWO main things.

(1) Push back against these “AI Safety” committee proposals.

These might looks like safety committees, but when you dig a little deeper, you realise they are speech and thought regulators.

(2) Build alternatives. Now.

Build many and open source them. The sooner we do this, and the sooner they can run more locally, the better chance we have to avoid a world in which everything trends toward homogenisation.

If we do this, we can have a world with real diversity - not the woke kind of bullshit. I mean diversity of thought, diversity of ideas, diversity of viewpoints and a true marketplace of ideas.

An Idea-Versity. What the original promise of the internet was. And not bound by the low bandwidth of typing and tapping. Couple that with Bitcoin, the internet of money, and you have the ingredients for a bright new future.

This is what the team and I have been experimenting with since the beginning of the year. Building smaller, narrow models that people can use as substitutes to these generalised LLMs.

We are going to open source everything we’ve done, and in time aim to make the best models compact enough to run locally on your own machines, while retaining a degree of depth, character and unique bias for use when and where you need it most.

I light-announced our first model at BTCPrague this week. The Spirit of Satoshi.

The goal is to make it the go-to model for a topic and industry I hold very dear to my heart. Bitcoin. Think of it as a globally accessible repository of all the best Bitcoin, Austrian Economics and Libertarian-like literature, made available through a LUI.

I believe it’s here that we must start to build a suite of alternative Ai models and tools.

I will write a dedicated blog on this next week, but until then, check it out, join the waitlist and reach out to me if you’re interested in helping us bring this to the world.

Until then.

Svetski