A.I. and the Future of Computing
ⓘ Notes are short, informal thought pieces relevant to Radiant Computer.
Language is the closest thing we have to thought, when it comes to mediums of expression. It’s the highest leverage tool we have. It’s no wonder we use language to program and control computers. Language is terse, direct and communicates intent. The command line interface (CLI) on UNIX-based systems embodies this perfectly.
But the command line can be opaque, hard to master, inflexible, inaccessible and ultimately inefficient for tasks that are hard to describe in words.
This is where LLMs shine: they can interpret language – often ambiguous – through context, providing a natural language interface to the computer. The UNIX philosophy tried to approximate this with single-purpose tools and composition via pipes, but ultimately failed to achieve it, as even experienced users struggle to string together complex combinations of tools into “sentences”. LLMs are in many ways the fulfillment of that vision: a natural evolution of the CLI.
> cat /var/log/syslog | grep "error" | sort | uniq -c | sort -rn | head -20
With LLMs we can finally have a computer that does what we mean. A computer we can program in a language we are fluent in.
The next evolution of the personal computer will bring everyone back to the command line, but this won’t be a VT100-style interface, it’ll be one that can natively render images, interactive graphics, sound; and output anything computable. A universal Turing machine where all inputs are permitted, and all outputs are possible. This living computational “fabric” will be able to represent anything and take any shape.
This is the ultimate outcome: no apps, no desktop, just a blinking cursor – a command line that puts all of computation at your fingertips.
Malleable Environments
As illustrated beautifully in Malleable Software, malleability of our environment is a key to creativity. We shape our spaces so that they can shape our work. This is how we should be thinking about computers: you start with a task, an idea, a goal, and the computer reshapes itself around the work.
For this to work safely, we can draw a boundary between the operating system kernel, which is fixed, and the environment, which is mutable and emergent. The kernel has one key responsibility: keeping all the moving parts of the system in check. In other words, ensuring that the permissions granted to the different parts of the system are respected. The kernel hands over specific capabilities to allow the parts to perform their role, and nothing else.
But we need not limit ourselves to software: with FPGAs, we have another kind of fabric; one that exists at the hardware level and can be reshaped just like software. The hardware/software dichotomy is gone: your computer re-programs its electronic circuits on the fly, based on the task.
Sovereignty
We believe that you should be able to fully own your computer. This means no dependency on third party services. It means privacy and control over your data. So how does this work with AI? Open-source models that run on commodity hardware exist, but they are either extremely resource hungry, or incapable of performing complex tasks.
If you want to run a model locally, and especially on mobile hardware, you need a small, specialized model. But specialized models are not that useful unless the system they serve and the environment they inhabit are themselves simple. This is where designing from scratch has an advantage.
A computer system built around a single language, with direct hardware access, comprehensible by a single person, would require a fraction of the knowledge to fully describe. There’s no need to train on hundreds of programming languages, frameworks, libraries and APIs. A model trained for this system alone could in theory perform better, with higher token efficiency and lower resource requirements. This is the only way to have both sovereignty and productivity with LLMs.
Conclusion
> |
It all starts with a blinking cursor. Instead of launching apps, you describe intent, and the computing environment assembles itself around you.