Stochastic Path

A New Start in 2025

2025-03-17T14:14:12+00:00

Suddenly it is the year of Ji-Si. There is a great deal to do this year and the goals are very clear, so it is certain to be a productive one.

The Warp and Weft of Time

The tide at the pen’s tip floods twelve rings of moon

ink-marks grow tree-rings in virtual soil

talking by the fire, stars fall into the teacup

ripples of drinking under the moon push open a spring window

fingertips weave the warp and weft of light

a metal forest grows inside the code

the galaxy whispers as it crosses the microscope

the formula of all things spreads its branches in the morning mist

pine-wind on seven strings wakes the sleeping grain of the wood

oil paint and rice paper elope into the dusk

when the thrown card cuts open the prism of light

dew is condensing on the arc of the taiji

starlight in the palm leads a child’s footsteps

the game is laid out under the eaves through four seasons

we practise measuring ridgelines with clouds

until the bearing of pine and cypress soaks into every bone

moonlight falls across the blank page still to come

grains of gold rustle in the hourglass

every unnamed seed has already set out

sprouting quietly along the line between dawn and dusk

Reading: The Power of Trees

2023-01-27T05:21:50+00:00

Before the holiday I read a small book: The Power of Trees, written by Gretchen C. Daily with photographs by Charles J. Katz Jr. I picked up a number of small facts from it — the limits on how tall a tree can grow, for one. The one I found most interesting concerns the largest living organism on Earth: Pando.

The largest such stand sits in Fishlake National Forest, in south-central Utah. Pando is a grove of aspen — Populus tremuloides. What makes it remarkable is that every tree is male, and the entire grove grew from a single tree. Besides propagating by seed, this aspen can also sprout from the parent root system. The Pando grove at Fish Lake covers more than forty hectares and weighs over sixty thousand tonnes, all of it joined into one by the root system underground — which makes it, arguably, the largest living thing in the world.

Reference: Sohu

A Few Flags for 2023

2023-01-02T05:21:50+00:00

The last post was my review of 2022. As always, here are a few challenging targets for 2023. There are also things in life that simply have to be faced; those do not go on the list.

Reading (15–50 books)
Broadening the view (6 new cities, and write the travel notes)
Certificates (driving licence, professional exams)
Interests (6–20 paintings, skateboarding)
Health (BMI into the normal range; every weight-related condition gone)
Writing (5–10 papers, 50 articles, publish a collection of drawings)

——

Paper of the day: REVIEW: Proteasome System of Protein Degradation and Processing — on the ubiquitin–proteasome system: proteasome structure, the ubiquitination machinery, the classical ATP/Ub-dependent mechanism of protein degradation, alternative mechanisms of proteasomal degradation and processing, and the damage to the proteasome system that leads to various diseases.

A Short Review of 2022

2022-12-31T05:21:50+00:00

In 2022 I set a personal record: the longest continuous stretch in a single city since I started university. For one reason and another it felt as though a great many things could not be done — and yet a great many things happened.

Around the office, most of the restaurants either closed or changed hands, over and over. The good news is that a few roadside places I like held on. Several friends’ companies expanded fast in 2021 and then had to face large layoffs in

And one after another I kept hearing that people I had read about in books and seen on television since childhood had died. In a year that so often fell silent, that news stood out sharply. Some things changed for me too, by choice and otherwise: more chances to sit still and go deep on the technical work, and to think about creating value, and about the variables in a life.

The work

I spent part of 2022 starting some genuinely interesting work. On the whole — value produced, research results — call it 80 out of 100. Some projects slipped, though I believe several important results will land in the first quarter of

As always, it is a privilege to work with a group of people who are single-minded about doing excellent work. It was a real shame that some colleagues left the team this year; I hope everything goes well for them in their new fields.

To do this new work properly I read a lot of papers across algorithms, biology, chemistry, materials and engineering, and I do think I learned something. I rethought where and in what form algorithms — data-driven AI methods in particular — actually get to be applied. Generative models made enormous progress in 2022: in painting, in conversation, AIGC gave people an explosively large imaginative space. If you think carefully about how AI technology produces commercial value, then rather than saying the best application is wherever large volumes of data are generated, it may be that AI has more room to land where the value of the data is underestimated.

I spent some time specifically on how to amplify the value of algorithms in biology. The main thing I took away is: go with the grain. The complexity of biological systems makes it very hard for us to design an entirely new system ourselves. The better approach is to start from the data, from the standpoint of the biological entity itself, build models that actually guide the work, and develop against the rules biology already follows. Optimising on top of an existing system is a far more efficient route from technology to result.

Finally I spent part of the year thinking about nonlinear growth — mainly, how to find the crossing points of a core technology between fields that look different but are in fact related, and how to design effective mechanisms for exchange and collaboration that amplify the value released by trading technology and data, so the whole grows quickly.

On the organisational side I tried IPD, gamification and so on, and read a number of management books — only to find that at my current half-understanding level, last year’s economist-exam textbook suited me best. So I gathered up some classic MBA texts and started reading. There is a line in Dalio’s Principles that is quite right: we fall into these holes only because we do not know the history. A great many things simply recur, and if you study history seriously you can know more of them.

By the end of 2022 I gradually felt I understood more about concepts I had assumed I already had a grip on — and respected them more. Innovation, for instance. Focus. The relationship between social impact and commercial translation. Financial models. A lot of things have to be thought about from a relaxed state; the thinking is like water held in your palms — you have to watch it quietly, and if you are tense and grip too hard, no idea is left at all.

For 2023 I want to take the application directions I can now see and extend them according to my own understanding, find their focal point, and concentrate on the core technology. I will try to pay closer attention to landing applications, and under a goal of long-term social value, build richer means and paths that reach long-term goals through short-term results. I tried building some processes in 2022 but understand them only superficially. In 2023 I want to concentrate on understanding one or two processes from zero to one, and see whether I can draw out any regularities.

Personal interests

I picked up a few skill points here and there this year. On the guqin I got as far as Ou Lu Wang Ji. On the skateboard I can just about land a fakie 180 — mostly a matter of not being bold enough, or I might have learned more. I gradually got the drawing habit back, did some pencil studies and tried oils. I did not hit the reading target I set at the start of the year, which may have something to do with travelling less. I read an assortment of unrelated books. In the first half of the year I finally finished several books on the Yijing that I had never managed to get through. When bored I did some research into the skeleton of the crucian carp; I may write that up later.

In games, besides continuing to explore Hyrule, the one that stayed with me was playing six or seven endings of Detroit: Become Human — including nuking Detroit. I finally reached the end of Uncharted 4, and played A Way Out. Death Stranding, Beyond, Nioh (punishing), the FFVII remake and Shadow of the Colossus are all still in progress. But the deepest impression was Edith Finch: the fates of the family members, one after another, seem to keep restating the difference between one individual existence and another, and the hollowness of a universal set of values. That may mean rather more given how 2022 went.

In the last month of 2022 I started learning Spanish; I can now ask directions, buy a ticket and order a meal. Also, I am better at lockpicking than I was in 2021, and I replaced both the inner tube and the tyre on my bicycle myself.

Finally

Some things happened in 2022 that had a large impact and about which there was nothing to be done. Those I will digest quietly, in my own time. Looking back at what I expected at the start of the year, a great deal happened that I had not expected. On the whole there were more shocks than pleasant surprises. Still, it is 2023 now. I hope everything gets better in 2023, and that we become braver.

Revive!

2022-03-26T05:21:50+00:00

Try to make use of github to host again the personal website. It looks like Jekyll is by default supported.

Jekyll also offers powerful support for code snippets:

def print_hi(name)
  puts "Hi, #{name}"
end
print_hi('Tom')
#=> prints 'Hi, Tom' to STDOUT.

Check out the Jekyll docs for more info on how to get the most out of Jekyll. File all bugs/feature requests at Jekyll’s GitHub repo. If you have questions, you can ask them on Jekyll Talk.

What AI Taught Me About Management

2019-03-09T22:49:42+00:00

Towards the end of 2017, at the invitation of someone well known on Zhihu, I shared some thoughts with a group about how a machine-learning way of thinking might help us. If what I said then was a reflection on how the specific habits of thought and behaviour in that field had changed my own thinking, then after a year or so of trying things and gathering practical experience, I have taken rather more from the frameworks of machine learning than from its habits.

What I want to talk about here is how the three elements of a machine-learning project map onto ordinary work and life.

In a normal machine-learning project, before anything begins, we usually have to settle three things:

The objective. What is the loss function — or, what is the function we want to optimise, and to what point do we optimise it?

The model. What kind of model do we choose; which cases do we run experiments on?

The search strategy. By what method do we find the “best” optimisation. “Best” here covers both whether we can find the optimal parameters and the cost or efficiency of the search itself.

Think about the tasks in your own job, and don’t these three words sound extremely familiar?

The objective. Why are we doing this, what result do we want, and can that result be expressed quantitatively? Thinking about the objective matters because desire is unlimited and resources — time, money — are not. When we invest in a stock, prepare for a negotiation, try a new line of research, run a marketing campaign or organise an event, the objective had best be explicit, simple and measurable.

Why think carefully about it? Because it determines whether we can review effectively afterwards — if we do not even know what we were trying to achieve, how do we conduct the review? And more importantly, it determines when we stop. Digging carefully into the objective and finding the most fundamental one keeps us from getting excessively tangled in some intermediate result; and when we do get stuck at an intermediate stage, a clear final objective helps us find a great many alternatives. Say we want to enter an unfamiliar market and hope to reach an agreement with company A quickly in order to gain a competitive position there. Then during the negotiation with A, we probably will not tie ourselves in knots over how the benefits are divided.

The model. To reach the objective, what kind of model do we want to use? For a new line of business, whether the core is R&D, engineering deployment, distribution channels or something else entirely dictates a different style of management. For personal development: a course, self-study, or a study group? If self-study, is it video, books, or scrolling Zhihu? In practice we usually implement the first solution we find. Thinking carefully about your model in advance lets you properly evaluate the options actually available to you, so that when one road turns out to be blocked, you can genuinely plan your way out.

The search method. When we have several models that might reach the objective, how do we find the most suitable one? From the automated machine-learning point of view we can do grid search — trying them one by one down a list; random search — trying them at random; or heuristic search — choosing the next candidate based on feedback from the last attempt. In real matters, which one you choose, and which “model” you try first, is usually shaped by the nature of the problem and the resources at hand.

Take the market-expansion example again. If the objective is to develop new technology on the basis of an existing market so as to raise the company’s influence and competitiveness, we might imagine these options: buy existing technology, develop it ourselves, or develop it jointly. Which one we pick sends us back a step to look at the objective again. Over what period does it need to be achieved? Do we have requirements on ownership of the technology? What quality do we need from the new technology, and can existing tools already reach it? If time is short and independence is not a strong requirement, buying may be a good choice. Conversely, if we want to build a higher technical moat, or if no good existing work on that technology exists, developing it ourselves may be the good option.

So in terms of search method: for the first case we would probably search existing solutions first and put the weight of the work into the partnership or purchase negotiation; for the second we would probably first assemble the research capability and assess the feasibility of doing it ourselves. And of course, if the finances allow, running both in “parallel search” is also an option.

The Pragmatic Programmer has a line I think is quite right: there is no perfect project. So when we consider a project’s objective we must fold the quality requirement into it as well — the completion date, the final quality standard, and so on.

Step back one more time. If the objective is to raise influence and competitiveness, is developing new technology the only “model” that reaches it? Would providing a better service do? Whether we buy or build, we may hit an obstacle we cannot get over. If our list of models has “provide a better service” on it as an option, I am confident we will be far more composed when we run into difficulty.

These three are best written out on a sheet before starting anything, thought through, discussed and listed one by one. Unfortunately, in a great many cases we begin with only a vague idea of why we are doing something and how, and only double back to think it through and fill in the gaps once we hit a problem. I am certainly not suggesting you freeze in place in order to complete a framework. But give yourself a quiet interval and work through what you are about to do along these lines. It may not be comprehensive — that is fine. Having gone through it, not only will the work go more smoothly, you will also spare yourself a lot of unnecessary anxiety. Filling in blanks is always easier than writing the essay.

Made By It, Undone By It?

2018-11-27T21:40:49+00:00

A couple of weeks ago the debate show Qipa Shuo ran a motion: if there were a technology that could, with one press of a button, synchronise all knowledge to every person in the world at once, should we support it? One argument from the opposing side was that because it reaches everyone, it would be terrible if bad people came into possession of that knowledge. The proposing side answered: we cannot tell good people from bad in advance, and if the worry is that bad people getting knowledge will harm society, then should we not have universal education either?

Yesterday’s news made me understand how deep that motion really is. Neither support nor opposition is quite adequate, because the real world is not a black-and-white division — it is a transformation from quantitative change into qualitative change.

One press to enlightenment? One cut to become a demon?

Some people are radical, some conservative. As we keep raising the quality of education, as more and more people can reach the most advanced science and technology, as more and more scientists and engineers make technology accessible so that it can be done conveniently in a small laboratory — outliers become that much easier to produce. At the Battle of Red Cliffs, Cao Cao’s army chained its warships together to withstand the wind and waves; had they been broken into small boats, some would certainly have been fast and some would certainly have capsized.

Twenty years ago we were saying that the twenty-first century would be the century of the life sciences, and a great many excellent people poured into the field — which later produced the joke:

Mathematics is the fire that lights the lamp of physics; physics is the lamp that lights the road of chemistry; chemistry is the road that leads to the pit of biology.

Twenty years on, immunotherapy, gene editing and other new techniques have entered public view — not only because they are new, but because they genuinely bring new hope for treating disease and for the long-term security of human life. We keep wanting to lower the difficulty of running an experiment and raise its certainty, and we want more people to take part in the research. We do not control participation the way nuclear testing was controlled decades ago; we hope instead to limit the use of these technologies through regulation and standards — much as, learning the lesson of thalidomide, we hope to control the risk of a new technology through rigorous clinical trials. And at the same time, we encourage more people to join in developing it.

But the larger the wood, the more kinds of bird. Once a technology can be carried out in a small laboratory, breaching the so-called boundary becomes far more likely. Technology, and the democratisation of technology, is a direction everyone seems to agree on and work towards — and yet today we do not appear to see clearly what long-term effects that direction can bring.

Last night, my friends’ concern for humanity’s long-term safety seemed to outweigh most short-term problems. Human society, the world around us, has never been perfect. Far too many people suffer enormous losses through the failings of others. Every morning there are traffic accidents caused by recklessness; we always hear that some factory has exploded, or that armed conflict has broken out somewhere; even the economic crisis that arrives every few years brings unrecoverable disaster to who knows how many. We worry about these things, we condemn them, we argue about them — but we seem to believe they are local, controllable, and will eventually end. Only this time the effect is permanent, the outcome unknown and uncontrollable. Pandora’s box has been opened.

Pandora’s box was opened long ago.

Its reverse side is greed and ignorance; its front is curiosity and fearlessness. When we press forward down the road of curiosity, we cannot truly strip away our own reverse side. The long-term impact of today’s science and technology may already be well beyond what humans can predict. We used to believe a machine would need twenty years to beat humans at Go; AlphaGo’s arrival in 2016 set off a fierce public argument about whether machines would replace people. Today, a news item about the birth of “Lulu” and “Nana” has shaken the research world and the world outside it alike.

Why did we not effectively prevent it beforehand? Perhaps because we always believe unconditionally that people are conservative and good — even though history keeps telling us that when 99% of people are, the 1% of exceptions will always bring us a surprise, or an outrage. And on the other side, we cannot accurately estimate the impact of technology; our understanding of society, a nonlinear complex system, is nowhere near sufficient. Most people cannot see the essence of today’s most advanced technology, even as we hear about artificial intelligence, quantum communication and gene editing every day. We cannot accurately foresee the effects they will have in five or ten years, good or bad — because we do not understand them. And on the other side again, even the technical experts cannot accurately estimate what the spread and application of a technology will do to society.

So what is the solution? It is hard to say. But at the least, I hope we can have as correct an understanding of the positive and negative sides of our own work as we did in the thalidomide case decades ago — though this time it is likely to be more complicated. I hope each party can offer a valuable judgement from its own professional standpoint, and eventually a solution.

This year Dying to Survive drew a wave of attention: patients buying drugs through non-compliant channels in order to survive and get treatment. A few days ago someone told me that a drug a friend of his is researching, though not yet approved, is already being used to treat some cancer patients. I have even heard of people studying treatments at home in order to treat a relative. I am not commenting on whether these stories are true, nor calling any single event purely good or purely bad. One face is the Buddha’s, the other the demon’s. What we need to think about is where, in these cases, the line between good and harm actually lies. Think that through, and we can go further.

Ethics is a question we would like to take seriously, and we do not want our own recklessness to endanger humanity as a whole. We also want technology to develop faster. And on yet another side, to borrow the title of an essay by Cao Zheng: do not challenge human nature. So how do we balance the three? The birth of gene-edited babies has been carried into public view, perhaps because the person responsible misjudged what the event would mean. That is a good thing. But if we really care, we need to think further ahead — otherwise technological progress plus natural fluctuation may bring us a good many more “surprises”.

On Evolution, and the Evolution of Intelligence

2018-10-13T13:07:48+00:00

During my postdoc at MIT in 2013, one of the pieces of research that stayed with me most was the work of Jeremy England, then a young professor there, who was trying to understand the origin of biological function at the level of physics. Put simply: if we take evolution to explain how species appear and disappear, then at the very beginning — as the world turned from the inorganic to the organic — what principles, what mechanisms, and what interactions with external forces shifted the tendencies of chemical reactions and gave rise to matter with biological function?

By biological function I mean the behaviours that, in the world as it now is, distinguish living matter from ordinary matter. On his university page England explained it like this: a cat falls under gravity, and so does a stone, so falling is not a life-related function. Against that, reproduction, the firing of a neuron, extracting energy from food, using the past and present to predict the future — these plainly have the colour of life about them. Starting from systems far from equilibrium, England explored how a randomly connected network of chemical reactions might, through self-organisation, maximise the efficiency with which it extracts energy from its environment, and used theory and simulation to ask how life arises in the world.

Leaving aside how far the theory is accepted in the field, the research poses a good question: how do we study living systems by the methods of physics? Are there different evolutionary paths, and how would we see them?

That is the end of the serious part. Everything below is one person’s speculation in the after-lunch stupor, with no scientific basis whatsoever. Read it as fantasy.

Evolution is a long process. If we treat the several billion years of the Earth as a single day, then the appearance and flourishing of humans is only the last second of it. We can identify the extinction of a species, but it seems very hard to observe a new one arriving by evolution. So it is entirely reasonable that simulation has become an important tool and method for studying it.

That enormous scale of time reminds me of the discussion Poincaré opens with in Dernières Pensées: if every constant in this world were changing in some way, would we be able to perceive the change? In a computer simulation of evolution, do the equations we put in guarantee that all the constants of nature involved exist in a fixed form? If some constant lies outside our knowledge — the way the speed of light in vacuum did a few centuries ago — then when we run a virtual evolution in a computer, is it possible that the equations governing that evolution violate, to some degree, the properties of those unknown constants, and cause them to evolve within the simulated timescale? If that evolution were slow, would we have any capacity to notice it? Or rather: to what extent would such a change alter, or fail to alter, our interpretation of the results?

Free interbreeding with no external interference tends, under the second law of thermodynamics, towards disorder, diversity and averaging out. Like a plume of smoke in a body of air, which always tends to spread outwards until it gradually disperses. The mixing of life, if nothing constrains it, should likewise develop in the direction of disorder — and so evolutionary theory brings in external energy to account for the appearance of more organised, highly intelligent living things.

If we take intelligence as an important index of evolution, is there a way of simulating that studies and discusses the evolution of intelligence alone? As Alex Pentland points out in Social Physics, one of the important objects of study in sociology is the flow of ideas — ideas move through social networks, change, gather, disperse. If an idea is an indirect reflection of intelligence, can that flow stand in for a kind of evolution of intelligence?

From the other direction, Antonio Damasio insists throughout Descartes’ Error: Emotion, Reason and the Human Brain that because there is no such thing as purely rational fear or judgement, the mind cannot exist apart from the material body of a living thing; a brain in a vat would have no mind. So would a purely computational study of the evolution of intelligence drift away from natural reality? Or how might one combine a real physical body with the study of pure algorithms?

Lately every forum, publication and gathering — formal and informal, official and otherwise — is asking the same question: where AI development is going, and what help or threat it represents for humanity. Set aside that, technically, the threat may still be very distant. If it did come about, would an AI evolve? And would the evolution of an AI be a counterexample to our rejection of Cartesian dualism?

I have gone on at this length because I have been planning to use some enjoyable methods to look, on a computer and quite unacademically, at the process of evolution and at the patterns by which new things and new behaviours appear — with a view to answering some of the questions above. If any of this interests you, get in touch. Perhaps we can start with a flow of ideas.

Winning, Losing, and the Original Mind

2017-05-25T22:31:33+00:00

This is a superb age.

I cannot imagine, or remember, the arena of thirty years ago — how many people put down what they were doing to watch, how many cheered for a victory. Today there is no victory, and there is no cheering. And yet as the position slowly grows complicated, as the sound of stones being placed grows louder, I think I can feel the same tremor that has run through people from antiquity onwards whenever a famous game took shape in front of them. A thousand-year-old game, carrying the desolation of winning and losing again and again, has walked all the way from the age of bronze weapons to the internet. In front of that small square, the people change, one after another — or perhaps they are only switching between zero and one. Strip all of it away, look only at the nineteen-line board, and the shock to the mind is not the slightest bit different.

At school a teacher told us that Romanticism is the constant leaping between two extremes. On the board, the exquisiteness of reason is exactly this dance on a knife’s edge: I do not advance a hair, and I do not retreat a hair. Black and white are tangled together, with nowhere to live and nowhere to die.

This is an age of breaking through.

The game has no boundary of its own; the people playing it do. That boundary was first a person, then a country, and now humanity. Within one country, the strongest player is the national hand. Perhaps at first the largest contest was only between the two people playing: you won, and that was that. Later the two sides came to stand for their organisations, their affiliations, their countries, their kind. We keep drawing ourselves into circles, the circles keep getting bigger, and the outcome keeps getting heavier. When we can no longer be told apart, we merge into a larger circle and challenge the next one. It is exactly through this loop — break out, merge, challenge — that we sharpen our craft, and that a game like this becomes something you can lose yourself in.

This is a noisy age.

Perhaps nothing is pure any more. Some hope for a human victory in the final match against the machine. Some mock how small human intelligence looks next to it. Some take the opportunity to sell a product with no apparent connection to any of it. None of this is right or wrong; we are already inside that turbid current ourselves. After all, beyond this game there is a great deal of life to get on with. But it is only at a moment like this — when we can ignore an opponent’s skin colour, ethnicity, even bodily form, and want nothing but to win, searching in agony for the divine move — that the mind throws off something magnificent. And for that I count myself lucky to be in this age, and to have witnessed a moment like this.

A passage I saw at a friend’s:

“Great Sage, where are you going?” “To tread the southern heavens, and shatter the highest firmament.” “And if you go and do not return —” “Then I go and do not return!”

After all, everything that comes from outside is beside the point. Winning and losing are a matter of the original mind.

Where It Gets Hardest

2017-03-08T22:16:48+00:00

Written on and off as the mood took me. Not necessarily right, and not necessarily wrong.

Setting aside all the things I am bad at — socialising, moving investors, and so on — the strongest feeling I have at the moment, and one of the hardest parts of building a company for someone like me, is how to restrain my own scattering ideas and keep the product’s scope inside something reasonable. Concretely, it comes down to the following, and I would very much like to discuss them.

1. Choosing a direction. This is the first thing and the most important thing. Talk to different people and you get different feedback. Is it SaaS, CRO, R&D, or a platform? Is it technology-driven, PR-driven, or is traffic king? Different users want different things: painstaking deep analysis, a comprehensive integrated solution, vertically targeted technical support, a general-purpose platform service. So what should we build, how, and on what schedule?

My effort here so far has gone into one question: who is the core user? My own definition of a core user is one you can reach quickly, who gives you a base to iterate and expand from, and who fits the technical character of the team. The first part means the technology you are researching cannot sit too far from the market — which requires understanding precisely what this group needs, and in what form they need it. The second means the market is large enough, and that your early users are the critical node in a complex network, the one hair that moves the whole body, or the entry point into the system. The last means the first two are things the team is good at and actually wants to do — otherwise you may as well go and find a good job. Once those are settled, features follow from needs, plans follow from the logic of the market, and everything else can wait.

2. Building the team. Given a fixed objective and the core team you already have, how do you expand and arrange people for the highest efficiency? Do you find partners with strong fundamentals and grow them, or bring in experienced people to survey the ground for you? Possibly neither is the best framing of the question. My answer now is that all of it rests on the analysis in point 1. Under the market and business plan that point 1 fixes: what kind of person do we need right now, what do we need in reserve for the next year, and for three years, and for five? Which colleagues are decisive in that process — people who can hold a whole area on their own, or who can open the door to future talent, whether by developing it or by recruiting it? Hire fast for what is urgent, and for the long-term core positions build a regular screening process so that the right people surface.

My own mistake before was this: whenever I met someone impressive, I wanted to find a way to work with them. I now understand more clearly that a startup has to be built around point 1. If an impressive person does not match the objective, they are not only a financial burden — they can pull the whole direction off course. Being friends is perfectly good too. This, I suppose, is what is meant by a thousand soldiers are easy to find; one general is not.

Of course, however NP-hard the problem is, there are excellent co-founders and there are friends holding all of this up, so every day is painful and happy at once.

Probably unfinished. I may write more.

If You Came to Interview With Us

2017-03-01T22:37:46+00:00

I have been looking for people continuously for a while now — found online, introduced by friends, machine-learning researchers, technical and non-technical roles — and I have spoken directly or indirectly with a great many of them. Here is the recruiter’s view of what tends to leave a poor impression in an interview. These are personal opinions; I hope they help, and I am happy to be argued with. (Everything below is void for anyone with an unusual enough background.)

First: the CV. For me and most candidates, the CV is the first exchange we have. So presenting one that is logically clear and plainly written really does seem essential to a good first impression. List education simply. Describe your professional background systematically. Where an experience stands out, explain it in a little more detail. Avoid long sentences wherever you can, and use the layout to separate the sections cleanly. I imagine you can read this advice in many places — but it genuinely, genuinely matters. Once it is written, show it to a friend with some visual taste and let them tidy it up. If you cannot find one, then check it yourself: consistent indentation and alignment, one typeface, no typos. As for a photograph, I personally think you can take it or leave it; a professionally styled portrait is a little strange.

Second: still the CV. Before you come in, go back and reread what you wrote on it. For anything within the last year, you should be able to state clearly what led to what and why. For anything within the last six months, you should be able to answer on the technical details. Your interviewer will certainly ask carefully about whatever you have highlighted, so prepare it properly. Otherwise you get the case where the answer to “what are you strongest at?” is “linear algebra”, followed by an inability to diagonalise a matrix. For my own part, I do not put much weight on thinking on your feet, because plenty of problems at work are ones I cannot solve on the spot either — I go and look things up. So when a candidate tells me they prepared specifically for this interview, I think that is a genuinely good sign.

Third: understand the role, and the company you might be joining. Some companies — ours among them — have very little material online, and a lot of the detail simply cannot be looked up. But if a candidate can tell me what they did manage to find, and bring specific questions about the company and about how the role relates to them, I am close to moved. How much you can say here, and how valuable your questions are, is genuinely a matter of what you put in: nobody can know us deeply, so the information you gather is more or less proportional to the time you spend. When I was job-hunting I would usually write a cover letter aimed at a particular company or role. Many positions no longer ask for one, but you should still prepare to the standard of being able to write it.

The rest really does come down to ability. The summary above is fairly general, but broadly:

A concise, logical CV.
Know yourself (be fluent in your own CV) and know the other side (understand the role you are interviewing for).

I have kept it general on purpose, so that you can hold it against your own situation and ask whether you have prepared enough.

We are also still looking for people with experience in computer-aided drug design (CADD) and Python engineering. Do get in touch.

Character Before Craft

2017-02-28T23:23:36+00:00

I was watching a programme in the evening and it quoted an old saying: before you learn to do things, learn to be a person; before you learn a craft, learn its ethics. I found I agreed with it completely.

There is nothing new in the line, which is why I have filed it under “old sayings, said again”. But for about a year now, every evening as I walk home down the same alley, the first half of it surfaces in my head — before you learn to do things, learn to be a person. Perhaps it is because I am always a little anxious in social settings, so I overthink after every event I attend. Having heard, and said, a great deal about how “we do business models”, “we do market segments”, “we do industry consolidation” — all these ways of doing — I find myself having to remind myself again and again, over all that noise: whatever you do, first be a decent person. My own reading of that comes down to two things. First, keep the desire to possess in check. Second, find your own line and defend it.

Everyone probably wants to possess something. Held badly, that want becomes the source of a great deal of unhappiness. Within a family, an excess of possessiveness can breach the last line of individual independence and produce resentment that seems impossible to explain. That familiar phrase — I’m doing this for your own good — is, to a large extent, “I” claiming rather too much ownership over you.

In business I have been revising myself continually. I began cautious, terrified of coming off worse in any exchange, and moved gradually towards open collaboration. When it is necessary, I am now willing to take the smaller half myself, as long as the objective gets met. The benefit of this is, first, that things actually get done quickly. Second, and more importantly, my energy stops going into fretting over gains and losses and goes into what matters. There is a third benefit: playing down the wish to own something lets me judge it from a more objective, more quantitative position. The player at the board is the one who cannot see it. Once you let go of treating a thing as your own flesh and blood, you can look at the position from outside it.

The coldness that the first principle produces has to be corrected, constantly, by the second. The hardest part is discovering your own line — or, put differently, your final objective. In building a company there are voices all around you. Without much experience of my own, I am one of the blind men with the elephant, and each voice describes some other part of the enormous animal in front of me. What I need is to find the best available answer under incomplete information. One way is to listen to more opinions and gather more viewpoints. The other is to work through the logic behind those different viewpoints and the relationships between them. There is a further difficulty: the optimum I am after should be a global one, so across time I also have to avoid one-sided judgements that trap me in a local optimum. And once you have found the objective, you have to hold it against every temptation. Like the foal crossing the river in the fable — having heard a great deal of advice, you still have to go and find out for yourself.

So the two halves of “being a decent person” while building something, as I understand it: find a suitable objective and a way of describing it, then go at it boldly — but never let yourself sit in my idea is unique for so long that self-admiration costs you your basic objectivity.

On Rational Deliberation

2016-07-03T13:17:12+00:00

This is the first piece in the series, so what should it be about? Let’s start with a behavioural test.

You have a box of fruit. It has been sitting there for a month and you only remembered it today. When you open it, more than half of the fruit is close to spoiling, and the rest does not look much better. You have two options (and of course we are thrifty people):

Eat the ones that are about to spoil. Nothing gets thrown away, but you may spend the next week eating fruit that hovers somewhere between good and bad — because by tomorrow the ones that already looked doubtful will have got worse.

Eat the good ones first. Some of today’s nearly spoiled fruit will inevitably have to go in the bin tomorrow, but at least the taste is guaranteed.

Which would you choose?

This is a well-known experimental design by the psychologist Fick Teister, and the result, in short, runs like this. Subjects who choose the first option tend to have a higher tolerance for the status quo and are more tenacious in their attachments. But because they may cherish everything they have ever owned a little too much, they never quite reach the summit of their own needs, and carry with them a permanent regret at never having stood on the peak to see all the smaller mountains below. Those who choose the second have far more decisiveness, but may act on impulse and miss opportunities. Who knows — perhaps the fruit that is about to spoil is still beautiful on the inside.

What I actually want to say today is this: it is remarkably easy to fabricate a test that looks like it holds up. To make it feel more real, the author can throw in dates, places, sample sizes, case studies. And most people, given the choice between forwarding and verifying, will — I suspect — lean towards the former, as long as it sounds like it makes a certain amount of sense.

So from this article onward, I sincerely hope we can treat this like a table of players in a game of Werewolf. For anything I discuss here — fiction aside — please argue from pure logic. Let us look together at the causes and the correlations behind an event, and not believe anything simply because an authority or a well-turned principle was invoked. On the same grounds, I have no patience for comments that come without reasons.

Consider this the opening. More will follow, slowly.