It is not uncommon for those in tech to see us (kind of) as crypto guys. For a long while, even we weren't sure whether our ‘positioning’ was most conducive to working with enterprise clients (or, might we say, them working with us), our doubts stemming from a fear of crypto that, it is commonly assumed, abounds in blue-chip companies. Hear this when we tell you it ought to be the other way around.
ChatGPT and Anthropic are looking into your prompts. The assumption that they will not look at them just because they are on American soil and these are American companies (and so Americans can sue them) is a false assumption. We back this up by asking one question, just one question: Companies often inform the user, following best practices, that their data is private. Assuring something completely stays private on an NVIDIA server takes literally one switch of a button—flip it, and nobody will be able to look into what you’re doing. Why then is it never advertised, never even so much as suggested by AI companies that they are doing everything to assure your data is yours, fully private and invisible to them? The only reasonable explanation is that it’s because they want control, masked by that patronizing assumption that you, the user, may do something harmful, and “we wouldn’t want that, now, would we?” So, in order to put some controls in place, they “need to take a look.”
When we talk about privacy, we are talking about actually ensuring nobody is able to see what those very same blue-chip companies we mentioned before are prompting from their AIs. This is an offer to customers who need this sort of control. We do not believe such companies actually voluntarily need a model provider’s eye over their employees or some tech that they themselves don’t want to install. They don’t need to rely on them to look at their data.
As a matter of fact, crypto companies are notoriously good at providing this privacy. They brought cryptography into the world, into a wide range of use cases. It’s actually because of blockchains that we have zero-knowledge proofs. Without them, this tech would never proliferate and would never have funding. Because of blockchain companies, we have cryptography in everyday life with wallets and self-custody. If there would not be blockchains, cryptography would be 20 years behind. Blockchains never control user funds; their founders get sued for this and go to prison for this. Remember the Tornado case? But it’s on blockchains that your data is safest—more than with any other company. There has been no single data breach in blockchain history, because there is nothing to breach; because of strong cryptography. Even as the tech industry relies on companies that get hacked right and left every day, many in it still have the gall to call crypto risky.
Privacy is not a sexy issue in the tech community because so many businesses have to rely on user data in order to make their money. In fact, most of big tech’s profits in some way connect to using user data. There’s an explicit risk associated with breaking a sort of compromise that a lot of these businesses have made. While many can make the case for why privacy matters in a humanist sense—and we know this argument—the case that we feel still needs making is why privacy should matter even as attaining that privacy means putting at risk the economic model for the companies that have become the main drivers of the tech sector.
In the enterprise world, someone needs to be sueable; someone needs to be shown as showing remorse. But we are talking about OpenAI and Anthropic, and we’re talking about blue-chip companies. We’re not talking about companies that do, for example, defense systems; they are definitely not relying on user data. So why does privacy matter in this context? Because all a modern company has is its data—let’s repeat that: all a modern company has is its data—and so naturally, they definitely wouldn’t want anyone to access it. Thus, what they are facing with today’s AI is almost like a paradox.
Anthropic gives them an apocalyptic view: they ought to be afraid of putting their work in Chinese hands because the Chinese will steal their data. “Wait a second!” We yell. “There’s only one way to actually stop them from stealing your data, and that is through private compute.”
This Chinese arms race about AI versus America feels like a manufactured issue to maintain the standing of the companies that managed somehow to get a foothold in the American corporate AI space. So are we then fighting on the side of the CCP? Only inasmuch as we want open source in AI, and the national entity that is working for this is the Chinese government, not the American one. There are a lot of contradictions in that. In a way, of course, the Chinese will take your prompts if you allow them. But the whole point of the open source models is that they can compete without outsourcing everything; they can be deployed on a company’s own infrastructure, rather than requiring the company to send its data to an external AI provider.
The Chinese will train their models if you use their service. But when you are a corporation, they cannot exercise the same level of control as big AI companies can. In fact, right now they’re not concerned with that; they are concerned with just the usage of their models and getting into the AI race. Frankly, they’re not a threat and are (kind of) giving an alternative to the world. Now, if someone is an American patriot, they wouldn’t prefer the open source alternative to their beloved nation’s best company, which they think will be the next driver of the US economy. And the competition coming from… doesn’t matter who, China or not China… isn’t all that well liked in the least. But for us, at least at the moment, the Chinese approach is more closely allied with our way of thinking about technology: we said at first that we must continue pouring money into open source if we are to ensure a democratic internet, and we say that now about AI. And whether it’s China doing this, or the US, or anyone else, doesn’t change our view that doing this is, well, absolutely great.
Our vision is that AI is the electricity of the future. No one owns electricity. We don’t want to have leaders in it just like we don’t want leadership in electricity. We believe that everyone will arrive more or less at the same level. Sooner or later, we will all arrive at… whatever the final 5% of the top possible LLM intelligence will be. Potential will be exhausted, and electricity will be electricity. The advance of one model over another will be marginal, exactly like in electricity, where the difference is the velocity, the current, and how good the infrastructure is. You can power a computer from a socket almost everywhere in the world where sockets exist, and sockets exist almost everywhere in the world, more ubiquitous than the mobile networks they power.
So the real question that should be asked is: what does the world look like when AI is electricity? The truth is: nobody knows.
Certainly, it is a world where we have a lot of ways to use intelligence. The problem right now is we have intelligence now on some level, and what people are using it for is not very optimized. They’re using it for searching the internet, like how to prepare Eggs Benedict. Or maybe generating videos of… I dunno, President Zelensky wearing a swastika armband or some other such nonsense. But let’s put into perspective the cries of bad actors: maybe… 3% of the population (?) is using AI for nefarious purposes. It’s not a lot of people using generative stuff that AI gives them. They closed Sora, and they would not do that if that was so popular. People used it for like 10 minutes, generated a couple of videos, and stopped it completely. Right now we are using AI in narrow ways. For the future, it is up to us all to build the manifold applications that let people enhance their lives using intelligence, beyond making videos and Instagram photos or whatever other time-wasting uses are currently on offer.
There is then a further snag that many have noticed: by introducing AI features into every product, many software companies have only made their products worse. AI companies tell you what you can use AI for, but you actually can’t, because the results are oftentimes simply… awful. Almost across the board, as a rule and as a principle, users cannot rely on today’s AI to be anything but marginally more productive. Want to do better intellectual work? You’ll achieve better results when you’re not using AI. Are you researching with AI? You’re not really researching, because you’re not reading all of the material, and real research implies that you read the material and you actually know what you’re talking about. Today’s AI has its utility found mostly in the parts where shutting parts of the brain off isn’t a disadvantage. Needless to say, this is the absurdity of the new, for anything truly new is inevitably absurd, to say nothing of how new things are messy. And we need find a fix for this dissonance for AI to become electricity.
Of course, we believe it will be. Whatever AI will go to—whether by reaching 95% of LLMs’ potential, or by some other AI approach like world models—there is no doubt that it will happen. But even when it’s fixed, there will still be a lot of people who will not use AI as we will have it. With every year, that segment will shrink. People who today complain (very rightly!) about relying on hallucinating machines will find (thankfully) fewer hallucinations. For a while, there will be no tools and apps that can more than marginally improve the parts of work and life that are based on intelligence. And it’s to those places where technology will go.
We need to have a light. Electricity without a lamp is nothing. It is this lamp we seek to build. Is intelligence just something you put inside an apartment to turn on televisions, order pizzas, and plan travel itineraries? That’s not intelligence. That’s bullshit. And skeptics are right to question why this bullshit gets so much traction. But it is we—we who are just now being allowed to dream of a future we have missed for so long—who will have to come up with the ways to make it work. And if ‘make it work’ sounds vague, it’s because it’s one of those things that has yet to be named. One can approach this problem by think sort of like we did when we played as children, inventing what our minds can do. Now we have to invent all over again what this intelligence of the artificial sort could actually bring to the real world and beautify, enlarge, and assist people in their journeys across this funny thing called life.
great!