Day 4 — RISC-V, and everything I’ve thought about it

A journey through architecture, forgotten languages and questions that remain unanswered (for now)

There are days when you sit down at your computer and wonder how you ended up there.

Not in an existential sense — or perhaps I am, a little — but in a practical sense: why am I studying RISC-V? Why this and not something else? Why didn’t I stick with Python like everyone else, or with JavaScript as every ‘how to find a job in 2026’ article suggests?

The honest answer is that I don’t really know. But trying to piece it together is a start.

The starting point: Prolog

It all began, at least in part, with Prolog.

Yes, Prolog. That language which many programmers are familiar with by name — often from university textbooks — but which very few actually use. A declarative language, based on formal logic, a far cry from the imperative paradigm we are used to.

I’ve studied it. I’ve used it. And I have to say, it’s changed the way I think about problems.

In practice, Prolog is used almost exclusively for structured data management — in business contexts, large or small, where the logic of queries and data relationships lends itself well to its model. It is not a tool for building web applications, nor is it designed for mobile use or scientific computing.

But then I saw something on YouTube that made me slow down.

A video — I can’t even remember who made it — which discussed Prolog as a tool for AI. Not for using AI, but for generating it. For building formal reasoning systems, for representing knowledge in a structured way, for creating agents that reason according to rules rather than approximating statistical patterns.

It’s a fascinating idea. And it’s also, probably, an idea that belongs to another era.

It is 2026. Large language models have redefined what it means to ‘do AI’. Symbolic reasoning — something Prolog does exceptionally well — still exists, but it has become a niche field. Using it on its own to build something competitive today seems, quite frankly, anachronistic. Not impossible, but unlikely.

A busy year, and what remains

2025 was a noisy year.

Anyone in the tech world knows the drill: announcements, launches, hype, setbacks, corrections, and relaunches. Every month brought something new — larger models, different architectures, tools that promised to change everything and sometimes actually did.

Anyone who kept their eyes open noticed a great deal. Not just the headlines, but the subtler signs: how companies’ priorities are shifting, where funding is going, what job vacancies are looking for, and which programming languages are appearing in the most active repositories.

And amidst all this, something tangible has emerged — or rather, has continued to emerge: a growing interest in open, efficient, customisable hardware architectures. RISC-V is one such example.

Julia, and the impossible combinations

For some years now, one name has been cropping up time and again in discussions about scientific computing and AI: Julia.

A modern, fast language designed for numerical and scientific computing, with a syntax reminiscent of Python but with performance levels approaching those of C. It is not new – it has been around since 2012 – but it has matured, and more and more people are using it in contexts where previously only Python, MATLAB or R were used.

The question that occurred to me — and I admit this is mere speculation, not a definitive argument — is this: does it make sense to consider combining Julia, Python and Prolog to build or enhance an LLM?

The idea would be something like this:

• Python for infrastructure, data and integration with existing frameworks
• Julia for computationally intensive tasks, numerical computation and optimisation
• Prolog for symbolic reasoning, formal knowledge representation and explicit rules

In theory, it makes sense. In practice, I don’t yet have enough information to judge whether it’s feasible, whether anyone is already doing it, or whether it’s simply one of those ideas that seem brilliant at night but strange in the light of day.

I’ll say it straight out: these are just empty words, for now. But empty words sometimes come back with something in their hands.

And RISC-V, at last

So: how did I end up with RISC-V?

I think it’s the natural outcome of a journey that leads downwards — not in the sense of decline, but in the sense of depth. The more you study how things work on the surface, the more you want to understand what lies beneath.

RISC-V is an open, royalty-free Instruction Set Architecture designed around the principles of simplicity and modularity. It does not belong to any single company. It is not proprietary. You can study it, modify it and implement it.

In a world where hardware is becoming increasingly central – in edge devices, AI inference and custom chips – understanding how an architecture works at the instruction level is not merely an academic luxury. It is a skill that is gradually becoming essential.

Day 4. Still so much to learn.

This post is part of a series of daily reflections on an open-ended and haphazard learning journey. Day 4 of something that doesn’t yet know where it’s heading — and perhaps that’s precisely the point.