the coming five years

In my first year of undergraduate I kept a diary of sorts on grammarly. A strange place to do it, I admit, but it was accessible across my devices - a set of Windows laptops, a PC tower, and a smartphone - and had a pretty default font. This was less of a diary and more a place for me to write down strongly-negative emotions in a coherent fashion to better understand them. I’ve added to the collection of entries (each reads a bit like a short story) over the years, with the last entry being in late August 2022. The first - early August 2017. My last blog post was late April 2020 and this one is early (mid?) April 2026. Perhaps something about the months draws me to similar experiences.

I used to write a lot more fiction - both prose and poetry. When I was 17 or 16, I don’t quite remember, I challenged myself to produce a new piece of content for my girlfriend at the time (we were long-distance). In retrospect, it was a great exercise in coercing creative output in a controlled fashion. I miss writing, hence this post.

That was a bit of an aside. The final entry in my “diary” was a poem:

The bygone summer
Brought and reclaimed with its passing
Glimpses of your smile and laughter.

While I was buried amongst my books
Mulling over some evasive frequencies,
You came out with me on that warm day's walk
Amongst the spires towering out in the distance
Just beyond the reach of the river's slow trickle.

I don't recall what coffee you ordered and sipped -
What stuck firmly in my mind were your piercing eyes,
Those clean sharp lines drawn out from them,
Your white linen blouse and light Levis jeans,
A gold chain necklace, hoops with large pearls.

What stuck was that little bumping-up against me,
That you admitted might have been intentional;
I worried the foliage would stain your white blouse. 
I remember writing after how much of a shame it was
That you'd be leaving soon, and how for the first time
In a real long while those ever-so-cliché butterflies
Made a courtesy visit to my stomach. 

The bygone summer's promised thunder,
That never quite got to echo through,
Instead, electric, left me quite rather
Light of head when
I leaned in to kiss you.

It was about a girl I had met in Oxford during the summer at the end of my Master’s degree. It’s passable, I’m not sure I’ve written better poetry but I’ve not written worse.

My college, St Cross, a modern graduate constituent college of the University of Oxford founded in 1965, shares premises with Pusey House, a slightly older religious institution dating back to 1884. Pusey House celebrates High Mass - making it more Catholic than Anglican.

I came into faith while I was at Oxford. I didn’t give myself much time to think while I read my undergraduate at Cambridge. Aside from the degree and the pandemic, I rowed somewhat religiously (intentional use) which left very little mental or emotional energy to process what happened throughout my life at the time and to properly reflect on my (and others’) doings.

The summer following my undergraduate and the first term at Oxford were the first moments, I’d say in perhaps 3 years, that I could sit and think. I won’t go into more detail, but it was overwhelming to a degree. I don’t exactly remember when this particular next anecdote occured, but it wasn’t too long after my second matrculation - at most a month or two. I distinctly remember the first time I consciously and voluntarily kneeled before Someone. It was a grounding experience, and a humbling one.

A few people have asked me if I regret doing a PhD. Supposedly I could have already started and sold a company in the meantime, or perhaps made the world a better (or worse) place tangiably and at scale by some other means.

The question reminds me of a conversation I had with a good friend of mine at 11pm in the Polish highlands. We were discussing what is more meaningful - a personal feeling of accomplishment, or your achievements being recognised by a broader community (at whatever scope is relevant). I argued, successfully as far as I could tell, for the former. The interpretation is open, but I cited Percy Shelley’s Ozymandias as one of the salient points:

I met a traveller from an antique land,
Who said — “Two vast and trunkless legs of stone
Stand in the desert. . . . Near them, on the sand,
Half sunk a shattered visage lies, whose frown,
And wrinkled lip, and sneer of cold command,
Tell that its sculptor well those passions read
Which yet survive, stamped on these lifeless things,
The hand that mocked them, and the heart that fed;
And on the pedestal, these words appear:
My name is Ozymandias, King of Kings;
Look on my Works, ye Mighty, and despair!"
Nothing beside remains. Round the decay
Of that colossal Wreck, boundless and bare
The lone and level sands stretch far away.

The time afforded to me during my PhD - a period of around 3 years and 2 months if I’m being pedantic, and closer to 4 if I’m being generous - gave me incredible flexibility to think. See the above section on why this is important. I often explain it to people by using the Russian phraseologism “войти в тело” which usually refers to becoming physically stronger, to get better, to gain weight, or to gain strength. I use it in the more literary sense of becoming more comfortable with yourself as a person (including being in your own body).

Aside from the immeasurable amount of opportunities that the degree presented - both in terms of getting to know a myriad of interesting folk, and being able to direct affect so many people’s lives (primarily through my teaching and running University societies) - coming into my own self was quite possibly the best outcome of my early-to-mid 20s; and the PhD and the Cambridge environment allowed me to do so in a broad meaningful way.

My thesis - “Technical (originally: Practical) and societal implications of machine learning security” - is a relatively broad piece of work spanning 240 pages (with references and appendices) that covers a mix of technical, legal, and human aspects of modern security in a machine learning context. I talk about the implications of my research and observed trends on the future of society and scientific discovery, as well as the likely challenges we’ll be facing. It concludes with a paragraph beginning with “Looking forward, I am optimistic.”

Optimistic I may be, but that comes from (a) being informed (b) being realistic about what’s to come. Well, what is to come?

We are attempting to answer that in an upcoming paper - “Human labour will become a Veblen good in the age of AI” - the title of which answers the question firsthand. The premise can be roughly deduced from the work-in-progress abstract:

Generative and agentic AI systems reduce the marginal cost of routine
cognitive, creative, and coordinative work. We believe this will
cause advanced economies to, in addition to automating,
reorganise the value of human labour into a more barbell-shaped order:
with cheap synthetic production at scale on one end and scarce high-status
human labour on the other. In this transition, many forms of knowledge
work move toward synthetic abundance, while labour that remains
recognisably human-embodied, imperfect, and socially accountable
becomes more economically and symbolically valuable.

We break down our arguments in three parts. First, AI compresses the
value of standardised middle-tier industrial and white-collar
labour by making high-quality synthetic substitutes cheap, scalable,
and often good-enough. Second, this compression does not eliminate
demand for humans, but reallocates it toward forms of what we call
performative humanity: work valued precisely for its visible human
character. Third, the resulting social order is likely to become more
stratified: mass consumption becomes increasingly synthetic and
low-cost, while premium markets increasingly pay for verified human
taste and presence.

We do not claim that this outcome is inevitable or uniform across sectors.
Rather, we argue that it is a plausible and already emerging equilibrium
that should be taken seriously.  If realised, this view implies that
AI changes not only productivity and employment, but also status structures
and the social meaning of labour.

I conclude by drawing the reader’s attention to a concept known as “Disruptive innovation” that draws upon Clayton Christensen and Michael Raynor’s book “The Innovator’s Solution”. This is related, but somewhat parallel to the

To paraphrase, “Disruptive innovation happens when a product that begins as lower-performing but cheaper, simpler, or more convenient improves until it satisfies mainstream needs and then displaces stronger incumbents”.

In the context of ML, frontier models follow the sustaining innovation path pushing benchmarks upward. Those improvements matter a lot for a niche subset of the userbase, but we are arguably already well past the point of diminishing returns - for many practical tasks there is a point where extra capability stops mattering much. At the point where open-source models are “good enough” - which they are - we start to see disruptive innovation.

Even as a whole, real-world disruption due to AI is less likely to come from maximised intelligence and more from acceptable intelligence plus better economics. We’ll see a lot more “good enough” models take hold in the near future - these will be open-source models run by infrastructure companies.