Is Language The OS of the Future?
Key insights of today’s newsletter:
Tech leaders like Andrej Karpathy are predicting the emergence of an LLM-powered operating system (OS).
It’s envisioned as a new system built from the ground up, with the LLM as the OS, AI agents as applications, and natural language as the programming interface.
While this may sound revolutionary and futuristic to some, it’s not the first time someone came up with that idea.
↓ Go deeper (6 min read)
During his first big public appearance after his departure from OpenAI, at an event of Sequoia Capital, Andrej Karpathy said:
“Roughly speaking, the way things are happening, everyone is trying to build what I like to refer to as a LLM OS.”
For those who don’t know, OS stands for operating system. Your operating system functions as a bridge between your hardware (your computer, your phone) and your software (your apps, etc.). Basically, it’s the thing that makes all the other things work.
Karpathy already alluded to the idea of an LLM OS in a tweet of November last year. And recent product launches like the Rabbit R1, which I covered in an previous piece; the Open Interpreter project (which may have crossed your timeline); and the Humane AI Pin can be seen as early attempts at this.
They all share a similar approach: you don’t need to control your devices anymore, instead the AI can. You just have to ask.
Programming in plain English
What Karpathy and others envision goes one step further even. They imagine not a new kind of computer, but a new kind of computing: with the LLM as the OS, AI agents as applications and natural language as the programming interface.
This sentiment was echoed by NVIDIA’s CEO Jensen Huang, a few weeks ago, when he said kids no longer need to learn programming. He said that instead, AI will enable everyone to code in their own language.
While this might sound revolutionary and futuristic, it’s actually a rather old dream that’s being dusted off.
In the late 60’s there was a strong belief that once the proper way of communicating with the machines had been found, all programming ills would have been cured:
Wouldn’t it be nice, for instance, to have programs in almost plain English, so that ordinary people could write and read them? People tend to forget that “doing away with the programmer” was COBOL’s major original objective. Fifteen years later, 80% of the world’s programming force was absorbed by the use of COBOL.
This is an excerpt from a speech by famous Dutch computer scientist Edsger W. Dijkstra, delivered at the ACM, the premier software engineering forum, in the year 1984. That’s four decades ago.
Funny enough, COBOL is still being used today, but only to maintain existing applications. Most programs are being rewritten in more modern languages or replaced with other software.
Creating machines in our own image
That’s not to say that our newest attempt will fail. It may very well succeed. Programming in natural language could unlock something long lasting and profound — or turn out to be something that we decide to depart from in a decade or two. Either way, it seems to be the direction things are headed.
Important to note here is that the true believers in this endeavor see it as a path to a new kind of machine, a more powerful machine that is able to reason and think deeply, like a human can.
Be that as it may, I do worry these systems will be less… reliable. Security-wise, the easiest way to hack a computer is to hack its OS and the easiest way to hack the mind is using language. So if we decide to make language to our new OS, we better make sure it’s protected against the powers of lying and manipulation.
And that is if things go off the rails intentionally. Unintended, unanticipated behavior might prove to be an even bigger challenge. You see, bug fixing a program is easy when it consists of code. Code is rigid, it’s precise. But language is fuzzy, constantly changing, and open to interpretation. Fixing bugs in natural language sounds an awful lot like… therapy.
What it most certainly will do is stroke our ego’s, of that I’m sure, to have these machines modeled after us. God made man in his own image, but man made machine in his own image. Little gods we are.
Join the conversation 🗣
Leave comment with your thoughts or like this article if it resonated.
Get in touch 📥
Shoot me an email at jurgen@cdisglobal.com.
We used English pseudo code as an intermediate programming step 40 years ago. As you say, natural language is not only full of ambiguity, but potentially over verbose for machine application. Gosh, I wouldn't trust an LLM to run an OS. Nor am I clear why human users would want or need to get involved at OS level.
What is an OS? People use this term metaphorically more than literally in most of the commentary I've seen on this topic related to LLMs. Access to common services? Maybe. Manage device and software resources? Less likely anytime soon. Traditionally, an OS is a service and translation layer b/w the application and hardware layers. I think most people assume it will work between applications or between users and applications for orchestration and microservices access. Much more likely. Whether that fits the definition of OS is a different matter. 😀