David Khachaturov

the coming five years

2026-04-12T00:00:00+00:00

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.

Current State of Jubeat Emulation

2020-05-03T00:00:00+00:00

I’ve been very interested in Jubeat (latest version is jubeat festo, released in 2018 to commemorate the game’s 10th anniversary) ever since first playing it back in 2017 in London’s Las Vegas Arcade Soho. Unfortunately (or fortunately, for my time) since I moved to Cambridge, there are few opportunities for me to go and play.

The game is notoriously hard to translate into a PC-environment due to the the 4x4 grid and custom cabinet. Jubeat is also less well-known in the Western world, compared to games like Taiko no Tatsujin or Osu! Tatakae! Ouendan (known as Elite Beat Agents) both of which are emulated (and arguably superseded) by osu!. There are custom $400 USB controllers that require a separate 27” monitor.

Bemani Dumps

The best way to play Jubeat is to get a bemaniPC (or use a HDD dump though it’s all a bit hush-hush; a better English level guide can be found here) and do some hardware hacking-about. After all that trouble, you can get a setup that’s close to the real thing.

There are a few good private trackers for Bemani content. A good summary of the whole situation, as mysticrudnin put it here is: “The right way to get an invite is to become active in bemani communities until you are friends with someone who can give you one. […] I don’t know what you think sows is, but the bemani community in general is very tight knit. People mostly know each other. It’s not just a faceless group putting up pirated content. It’s close and real-life friends sharing with each other. People who come in to sows and snatch stuff without being a part of the community and contributing aren’t really welcome. This is both good and bad from your perspective. If you’re genuinely interested in bemani, you won’t find it hard to get an invite.”

Best way to get into this scene is to join a Discord and start chatting to people - it may be a bit imposing, but people are usually happy to help if you ask nicely enough.

If you’re intrigued (rightly so), an interesting article on Konami DRM was written by mon.im.

PC Emulators

There was once two main emulators that didn’t use dumps but memo files instead (explained later) :- Jubeat Analyser, and Youbeat (note the use of past tense and a link to WebArchive). There is also an “improved” (in quotation marks because I can’t personally vouch for this) version of Analyser called Jubeat Analyser Append (the download has been discontinued but you can easily find the MediaFire link via Wayback).

Memo files are just text files that encode the note timings, much like .osu files. The main two repositories for these are jubeat_memo and cosmos memo. They do not come with the song MP3s. Jubeat Analyser supports both file formats (I think?), but there are tools to work around that.

On a technical note, here is an explanation of the file format. It’s pretty understandable with the use of Google Translate. There is also an open-source Jubeat-Emulator which would be useful for understanding.

My Masterplan

I’m rocking an iPad 4 (perpetually on iOS 10.3.3) which doesn’t give me a lot to work with, but it’s a touchscreen and a relatively big one at that. It’s too much of a bother getting Jubeat Plus working on iOS and it also severely limits the song options. On the other hand, Jubeat Analyser is just grand but playing using a keyboard isn’t fun…

Enter the most overly-specific engineering problem to date. Use Rickpactor (because CydiaImpactor doesn’t work anymore, may it rest in peace) to load the iOS version of LÖVE onto the iPad. Write a quick and dirty LÖVE program to display a 4x4 grid and send network packets to my PC with relevant pressed/released data. Then write a quick C# feeder that will pick up these packed and map them to keystrokes (matching up Jubeat Analyser’s setup).

I’ll make another blog post if I get that convoluted approach working..

that convoluted approach (3/5/2020)

I got this whole thing working faster than I expected. LÖVE will never cease to amaze me with how great it is for prototyping. The GitHub repo can be found here. Have a look at the video below for a demo, keeping in mind I haven’t used the setup long enough to get good (it’s hard to know where the buttons are without looking and without haptic feedback; though I have an idea to fix that…):

On Motivation

2020-04-27T00:00:00+00:00

A few people over the years have asked me about what drives me, what inspires me to keep waking up in the morning and trying to do my best.

Unironically, the answer is often Boku no Hero Academia. The anime. If the MC can do his best every day and try to overcome his limitations, why can’t I? This has been a big inspiration from April 3rd 2016, when it started airing. This coincided with when I started watching it.

The show gained extra significance in my life when I started rowing during my first year of undergrad. The physical development aspect of the anime now directly coincided with my own goals. The ‘semantic gap’ (pardon the abuse), between my source of motivation and where I wanted to channel it, decreased.

It really helps that I am interested in the things that I want to be motivated in. For example - I genuinely enjoy rowing; and programming; and writing. If I was trying to get myself to do research into operational semantics, the story would be quite different. So perhaps the secret sauce is luck? I just got lucky that I am disciplined with my very wants.

I could attribute that to a roll of genetic dice, but that’s not interesting. There is definitely a sense of inherent masochism in everything that I do. Whenever I wanted a chocolate, but knew that I probably did not exercise enough to warrant it (from a calorie-balancing perspective), I would stop myself. These subtle “stop right there, sir; don’t do that now” moments came up multiple times a day. In a way, I was conditioning myself to want the right things (whatever my definition of right was/is).

Having goals is important, but often the best things in life come about organically. My most popular project, BetterJoy, was born just because someone on Reddit asked whether it was possible to use the gyroscope found in Nintendo Switch Pro Controllers as input to Cemu (a Wii U emulator). I figured “Why shouldn’t it be?” and decided to sink my teeth into it. From there on, the project began growing organically and I was motivated by the (mostly) positive community feedback and recognition that it got (and perhaps GitHub stars).

The concrete moment when I realised that I made something bigger than myself was when I found a random post of Baidu or something other Chinese-language site discussing my program. Through the liberal use of Google Translate, I managed to read through the forum thread and found that BetterJoy (then known as BetterJoyForCemu) was genuinely being used! There were people offering advice on how to overcome some technical bugs to other people, and even some forked versions of the program that supports off-brand controllers.

My motivation for rowing though, is not as organic. It was and continues to be a way of proving to myself what I am physically capable of. That, and all of my friends in first year took up rowing; I needed to get in on the banter. To quote my as-of-yet unreleased novel(la) “About Sport” (working title):

“Drive and creativity - what more does one need to propel themselves into their work? As long as you have your mind set on achieving something, anything is within your grasp. Even if your reach is far too limited now, stretching out, day by day, will see gradual improvement. Come a month, two, or even a year, you’ll look back and be surprised by how much further you can see, feel and touch.

“Sports have always appealed to me as a way of finding out where my maximum potential lay. To be able to feel every nook of your body swell and heave; to tame that screaming aching voice that wants a rest; that is superseding yourself. To cross the finish line, performing those last, almost mechanical motions that have got you that far in the first place, and experience the ever-blissful moment of calm.

“When the rushing and beating blood in your ears is no longer in a fight with your muscles, and the world is buried or rather masked by your internal cacophony that was there from the moment you started counting down your actions. That moment of ironic silence when nothing else exists other than your own body and the sweat running down your face; that moment is the reason to play sports.”

My next goal, with regards to rowing, is to trial and join a team at the Cambridge University Lightweight Rowing Club. Well, as of today (27/04/20) the University of Cambridge Boat Clubs (CUBC, CUWBC, CULRC) announced that they will be combining into Cambridge University Boat Club. So it would be more correct to say that I want to get into the CUBC’s men’s lightweight boat.

My training plan for that follows a modified version of The Pete Plan with more steady state™ and extra bodyweight circuits. This results in about 85km of rowing a week. There is a lot of discussion right now about high-volume training, but since I am nowhere near Olympic level, I think that will do.

What drives me? To see what the best “me” could look like, from a mental and physical perspective.