Edouard Fouché

What remains once AI overtakes most of the execution? Taste.

Mon, 20 Apr 2026 23:00:00 +0000

Taste is not whether one uses short rather than long sentences, or whether one dares to use the infamous em-dash with style.

Taste is using one’s best judgement to determine whether the execution is right: whether it fits the intention and original purpose. You can generate a thousand lines of code in minutes now. But deciding which of those lines belong in your product, and which do not — that is taste.

Interestingly, taste is also about setting the intention. Answering questions such as:

What is worth building?
What is worth writing about?
What is worth standing for?

Taste is unique to a particular person and develops over time. Still, taste tends to follow a consensus among a community of connaisseurs. There are wines most people agree are outstanding, yet they do not pair well with all kinds of food. There are general rules in writing (text, software) that most people consider acceptable, yet they do not fit all purposes.

In the future, we will not be looking for people with specific skills, e.g., the ability to write code, but rather for people with a taste for something specific, or the willingness to acquire such taste.

What is interesting about taste is that there is something very general about it: the process of forming an opinion, weighing the pros and cons of alternatives. And at the same time, something very specific: your taste in poetry will not help you much in software engineering.

I think that this Meme with Rick Rubin (which is going somewhat viral in the coder community) caries a deep misconception: While taste is indeed distinct from technical ability, practice (which in turn leads to some level of technical ability) is required to acquire taste.

Rank in your head the top five disciplines in which you have acquired the most taste. I bet at least one of them is not work-related, and I bet it ranks high.

Would you like to share that one with me? I’ll go first. Or, just let me know what you think about taste.

The Identity Crisis of Software Engineers

Mon, 13 Apr 2026 23:00:00 +0000

Something big and unsettling is happening in software engineering right now. Not dramatic enough to make headlines on its own, but persistent enough that I keep noticing it in conversations with colleagues, in forums, and increasingly in myself.

Software engineers are going through an identity crisis.

With AI coding assistants now a standard part of the workflow, many of us spend more time reviewing code than writing it. There are entire days where I write almost no code myself, and instead read hundreds of lines generated by an AI.

That sounds like a productivity gain. It is not that simple.

Reviewing AI-generated code is harder than writing code yourself. When you write, you hold the intent in your head. When you review, you have to reconstruct it, and then hunt for subtle bugs that a fluent but fundamentally pattern-matching system might have introduced. Eight hours of writing code is tiring. Eight hours of reviewing it is exhausting in a different, more draining way.

And yet, it still requires you to be very good at writing code. This is the paradox. Effective review demands deep expertise, arguably more than writing something yourself from scratch. Code review at this level has always been the domain of senior engineers, who have honed their skills over the years. AI has not removed that requirement; it has just moved it earlier and made it universal.

But many people entered this field because they love to write code. The satisfaction of cracking a hard problem, of watching something work because you designed and built it; that experience is increasingly replaced by a different kind of task. You do not solve the puzzle anymore. You verify the solution. That is not the same thing.

The lines-of-code metric used to be a flawed but real proxy for progress; something some engineers optimized for and felt proud of. It is now essentially meaningless.

There is also a profound shift in accountability. Before, when you reviewed code, the original author carried most of the responsibility for its behavior. Now the author is an AI that cannot be held accountable, which makes the reviewer the sole responsible party. More responsibility, less creative ownership.

Altogether, the profession has arguably become more tiring, more demanding, more accountable, and for many, less fun.

Personally, I find it more interesting. It pushes me toward questions I find more rewarding: not just how to build something, but what is actually worth building. That feels like a more human kind of problem.

If you are a software engineer, or aspire to become one: how do you see what is happening to our profession right now?

Anthropic's Claude Code Leak

Tue, 07 Apr 2026 23:00:00 +0000

Last week, Anthropic accidentally exposed internal Claude Code source through a packaging mistake, and that may lead to one of the most instructive AI product lessons of the year. Anthropic itself admitted this was a human error. Everyone can now discuss what this leak revealed, and there are at least a few important things to learn from it.

1. In AI, speed of shipping can become a liability if release discipline does not keep up.

Anthropic removed the affected version quickly, but the code had already spread. Once something like this is public, it is effectively impossible to contain. This was not the first time Anthropic leaked Claude Code source, which makes the incident even more striking.

2. What leaked was not just implementation detail; it was product thinking.

Public analyses suggest the exposed code revealed not only orchestration, memory, and context management patterns, but also references to hidden or unreleased features such as Buddy, Kairos, Dream, Ultraplan, and Undercover mode. That is far more valuable than a simple code dump. It gives competitors and researchers a window into how one of the best AI coding products is actually built.

3. This leak reinforces a point that many people still underestimate: the model is only part of the product.

The most important differentiation in LLM-based tools often comes from the harness around the model: orchestration, memory, context compression, tool use, permissioning, safety controls, and workflow design. The analysis of the Claude Code architecture strongly points in that direction. The real competitive advantage is often not the model alone, but the engineering system built around it. Claude Code source code is huge (500K lines of code over more than 2000 files).

Over the next weeks, we will probably see competitors absorb some of these ideas, close some feature gaps, and reproduce parts of the experience faster than they otherwise would have. Good for users. Less good for Anthropic.

And that leads to the uncomfortable question:

If the product can be studied, copied, and iterated on quickly, where does the moat of an AI product actually come from?

The underlying model?
Execution speed?
Product design?
Distribution?

Even if a company like Anthropic is able to keep reinventing the harness faster than others can copy it, we will eventually converge and reach a level of maturity from which marginal improvements will become less significant.

One more thing makes this especially interesting: Anthropic has positioned itself as one of the most careful and safety-conscious players in AI. So when a preventable packaging mistake exposes strategic product IP, it does not just raise engineering questions. It raises credibility questions.

The AI race is not about who has the best model anymore. It is about who can build, ship, protect, and continuously improve the best system around that model.

I am curious about what you think: Does this kind of leak materially weaken Anthropic’s moat? And of the four pillars above — model, execution speed, product design, distribution — which one do you think matters most right now?

Why using AI requires us to be even better (as developers and beyond)

Wed, 25 Mar 2026 23:00:00 +0000

In almost every movie involving air or space travel, the pilot relies on autopilot most of the time. But there always comes a heroic moment where they have to switch to manual mode and save the crew through sheer skill that almost no one else aboard possesses.

I have observed strikingly similar situations while debugging code with the best AI models available.

When a problem becomes slightly out of the ordinary, the AI copilot will quickly modify the codebase, form wrong hypotheses that reinforce themselves as the code evolves, and lead you into false beliefs about why things are not working. It takes real skill to hit pause, step back, and understand what actually happened, especially when you are already in unfamiliar territory: maybe using a library you have not touched in a while, maybe solving a problem with no obvious precedent. Resisting the confident, fluent misdirection of an LLM in that moment is genuinely harder than not using an LLM at all.

AI copilots are remarkable. But it is tempting to conclude: “We do not need a pilot anymore” — or worse, “We can get away with a less qualified one.” I see many decision-makers fall into that trap.

As long as what we are building involves any degree of innovation, that thinking is extremely risky.

The good news: knowledge workers will increasingly get to work on problems with higher degrees of innovation, because AI can do almost all of the non-innovative work. The question worth sitting with is: Are we cultivating the deeper skills that make that possible, or are we quietly letting them atrophy?

Have you experienced something similar? I would love to hear about it.

On becoming a German citizen (as a French native)

Thu, 28 Nov 2024 23:00:00 +0000

Last week, Nuremberg City Hall hosted a celebration for all its residents who obtained German citizenship this year, myself included. It was a nice opportunity to meet fellow “new” German citizens, and shake the hand of Mayor Marcus König.

I am now proud to call myself a French-German citizen of Europe. Despite our countries’ tumultuous history, we have managed to cultivate a unique bond of friendship. Growing up, I was exposed to the stories of my grandparents, who experienced World War II as children. My grandfather was unable to attend school because his village’s school was destroyed by German bombers. The Gestapo arrested the local priest for hiding Jews, and he was killed in a concentration camp.

Despite this painful past, I am pleased to represent the friendship between our two nations. It shows that within just two generations, the horrors of war can be put behind us, without forgetting the lessons they have taught us. I feel humbled to contribute to the legacy of those who have championed the French-German friendship, following in the footsteps of Adenauer, de Gaulle, Brandt, Pompidou, and others.

As an added bonus, I now get to vote twice more ;).

How about you? If you had been living in Germany (or any other country) for a long time, would you consider obtaining its citizenship?

Observations about my transition from Academia to Industry

Tue, 05 Dec 2023 23:00:00 +0000

I was invited to the “To be Honest” conference organized by the German Scholars Organization (GSO) to share my experience about my recent transition from Academia into the industry. In this blog post, I will share my first observations about my transition, which occurred only two months ago.

The session was called “Heartbreak or Dance of Joy? Careers After Academia.” The organizers kindly sent us a list of guiding questions to prepare for discussion. So, I will use those questions to structure this post :).

→ What is your current job about?

I am a Data Scientist for Product Innovation at Siemens Smart Infrastructure. My primary role is to develop and deploy data-driven products that are of general interest all across the company. Of course, I am not alone; I work with a team, but I mainly contribute with my expertise in Data Science, the field I specialized in during my studies, Ph.D., and Postdoc. I am also active on a less technical level in data strategy topics.

→ At what point in your academic career did you transition?

I have transitioned after three years of being a Postdoc (I have been employed at the university for a total of seven years). The position I had is sometimes called “Group Leader”, but I find that label misleading. I had lectures to give every week, exams to conduct, and four PhD students that I was directly advising (without being their official advisor on paper – but that is the usual way in Academia). I was also involved in administrative committees at the university. My work was similar to that of a Professor but without the corresponding title or recognition.

And already, I had a feeling that I had seen enough of the academic world. As you step up, you get to do less and less research. You must spend most of your time teaching, handling administrative tasks, reviewing papers (from your students and external), managing students and technical staff. That last point can quickly become very frustrating if you cannot choose (or poorly choose) your staff. Academia has a way to suck out the fun of research, which I was not aware of during the Ph.D. journey.

At this stage and beyond, not choosing your staff may happen for various reasons, e.g., some people are here already, and you have to deal with them; you cannot simply let them go. Or, you do not have your own lab yet, and the head of the lab does not involve you in hiring decisions.

As you progress in Academia, the expected research output increases (you are an experienced researcher, right? So you can write more papers), but actually you have less control over productivity. What you can achieve depends on your environment and factors that you cannot control. In my opinion, the most important aspects are: good students, enough funding, and a supportive mentor.

Until you become a Professor, you do not have much control over your environment, and even once you become Professor, your control is limited. You may have excellent research ideas, but if the research topics and your profile do not fit the current expectation of funding parties (which can be influenced by various trends: Blockchain, Generative AI and so on), you will likely not get the funds to perform your research. For example, I have met researchers who would artificially inject certain buzzwords into their project proposal to get the attention and approval of reviewers.

Ultimately, progress in Academia becomes a game of luck and possibly involves compromises on integrity. One pattern I see when looking at some successful researchers is that they tend to publish the same idea with a slight twist many times, but sell it each time as brand new. To some extent, staying in Academia is not survival of the fittest but survival of the luckiest and least scrupulous. Don’t get me wrong, at that level, everyone is excellent, and there are still outstanding researchers out there. But, the stakes are high, and the general attitude of “publish or perish” leads well-intentioned people to bend their integrity.

→ One thing that surprised you after you stepped beyond Academia?

Everything must go faster (time is indeed money). Still, we usually have better tools (money is not a problem, provided it brings you to your goal / the company’s goal).

In Academia, everyone is super-expert in their field, and even when you explain your science to someone from another field, you can expect enough science background from them so that they understand easily. Outside of Academia, you often have to step down the ivory tower and explain using two or three levels of abstration lower, and accept that it makes your explanation less precise.

Work/life balance is much better, though that last point depends on the company and your individual behavior too. Generally, how we approach time in Academia versus industry differs widely. Formally speaking, when you have a contract of employment, you sell your time (that holds both for industry and Academia). But, in Academia, we tend to think as if we “give” our time. We literally “give” our time in Academia for various reasons: passion for research, higher purpose, conference deadlines, (unreasonable) reviewing deadlines, teaching, mentoring, or the “number of papers” required to reach specific goals.

Also, compensation in Academia is usually not based on performance, but grade and length of service. It is easy to get locked into a grade (or maybe, you have reached the highest grade possible already?), and then you have little to no evolution. And if you suddenly decide to stop delivering, compensation will likely not change either.

In turn, it is often easier in the industry to quantify your impact on the business (that depends on your branch), which makes it easier to set a fair price on your time and to negotiate. If you feel your engagement and impact go beyond your current retributions, you should gather proof and ask for a raise. Ultimately, time is the only resource you cannot get back – making it the most valuable one. Some companies may try to get every drop of productivity out of you and may drive you to burnout. My general advice is to be aware and careful about that.

→ Heartbreak or dance of joy? How did it feel for you personally to move on to a career after Academia – and why?

It is a leap into the unknown, so that is scary. Also, leaving the academic freedom did trigger a bitter feeling. After all, it was nice to have all this time to pursue something one wanted to do. But life goes on! For me, leaving was more like a dance of joy. I was fed up with the academic system and did not feel I could continue to thrive there. I could still have become a Professor, but I thought: “I will become a Professor, and then what?”.

In the industry, career evolution is much more diverse, and I value that diversity. I also value that I am not so much tied to my employer. If, at some point, work becomes less enjoyable or my life situation changes for whatever reason, it is relatively easy to find another job somewhere else. That is not like that in Academia; unlimited employement contracts are so scarce that if you ever get one, more likely than not, you will stay in the same spot for your whole career.

When I decided to leave, I was afraid others would think (that alone, worrying about what others think, is a mistake) that I left Academia because “I was not good enough to become a Professor,” but I know for a fact that is untrue. I left because I wanted to avoid becoming a Professor, which is a different perspective. That “not-good-enough” thinking is based on the premise that Professors are where they are because they were the best or most promising researchers in their field. That is what many Professors believe, and it is not true! Most professors got the job because a committee (mainly composed of Professors, many of whom may have a hidden agenda) decided they were the most fitting candidates at this point in time.

I would advise anyone considering whether to leave Academia to think deeply about whether they want to become a Professor and to take the time to consider the positive and negative aspects of the job for themselves. If the negative aspects outweigh the positive ones, then you should leave. If not, going to the industry may also help you become a professor, ironically (more on that later).

→ What are the differences between working in Academia and your current job? (Or: what are similarities?)

Teamwork is much more important, and I feel much less isolated. I am also in a very large company, so there is always someone you can ask if you have a question about something. Quality expectations in the industry tend to be a bit lower; it must be “good enough” and not “perfect” because investing the time to make things 5% better is not worth it economically. In Academia, we tend to only focus on those 5% because we have the freedom to do so, and doing anything else would be perceived as not novel enough.

→ What makes PhD graduates and Postdocs well-equipped for careers beyond Academia?

We are extremely good at comparing things. This is a hard thing to do; it requires patience, courage, rigor, method, knowledge in the field, openness, and resistance to your own cognitive biases. People who have not been in science long enough tend to have difficulty doing that.

Also, we have learned to refrain from taking arguments at face value: “It is not because it is written in a book that is true”. Through the PhD experience, we have become more genuine, and we are not afraid to say “I do not know” and to speak up when something feels odd.

→ Was there anything you found challenging in transitioning – and if so, what helped you address those challenges?

One of the hardest things is explaining what I did during my time in Academia. I have the feeling most people do not understand what I did and have no clue about the process I went through.

And, we from Academia tend to largely undersell ourselves. Some people in the industry would talk about themself like they are expert at something, but the level they need to call themself “expert” at something is very far below the level we need as researchers to call ourselves an expert. As a research, you are a super-expert!

That is important to remember. And, if you have a PhD, feel free to brag a little about it…

→ What has helped you the most in getting to where you are right now in your career (e.g., skills, networking, visibility, mentors, etc.)?

A skill: Resilience. During my Ph.D., I have learned never to give up. Networking is beneficial, too. For example, having someone in your network working for the company you are applying to can be beneficial. You can contact them about interview preparation, salary negotiations, and when you are about to decide whether to join or not.

→ How can you prepare yourself for a career after Academia?

The best way is to do one or several internships. It will give you insights about how the industry works and create important relationships for your network. Try to break away from the stereotype of the crazy scientist and prepare to explain your research in super simple terms. You want to send the message: “Yes, I am a scientist, but I am so much more than that.”

I did not have the opportunity to do an internship during the PhD, but I did several internships before, which has been helpful.

→ Is there a good time to leave Academia? Or, in other words: Can it ever be too late?

It is never too late, but earlier is easier. After the Ph.D. it gets harder and harder. Still, there is a market for senior Professors who are hired to lead a corporate research lab, but that is much later. You first need to become somewhat famous, which is not guaranteed and may not be what you want. I think the best time to leave Academia is right after the Ph.D. I find that the industry does not highly value Postdoc experience. But if you had a Postdoc experience, it is good to use it to illustrate your leadership skills (you were not a “student” anymore, right?). I find that landing a job right out of Ph.D. is challenging because near the end one is usually too busy to engage seriously in job hunting.

→ Let’s be honest: what should we be aware of when it comes to leaving Academia?

Many positions are available, but finding a position that will be truly interesting to you is challenging. Many companies are looking for “good workhorses” and are thus not ready to hire a Ph.D. holder who will start by questioning the task first. The other problem is that companies might expect you to be more expensive (and you should be! Ph.D. time is work experience, and you have a great degree). So, why hire an expensive Ph.D. holder if they expect someone else to do the job just as well? In fact, doing the Ph.D. closes more doors than it opens, but many of the remaining jobs are more attractive (to me, at least). Even if you are fantastic, you will face many rejections, and it will take time to find a position. But you must understand that these rejections have nothing to do with you. Often, the position is being given internally already, or they are looking for a profile quite different from yours or whatever they say they are looking for. It is a game of patience and perseverance.

→ What advice would you give to your younger self (PhD candidate/ Postdoc)?

Do not worry so much. It is going to be all right. And push that idea of yours to do a Ph.D. internship. It will make your life easier later. Keep focusing on yourself rather than on what others think. You can only compare yourself with your earlier self. People tend to be much less brilliant than how they present themselves.

→ What piece of advice do you give to younger researchers who are considering transitioning beyond Academia today?

Many people in Academia will try to discourage you. Please do not listen to them and find out what you want to do. Consider what it will take from you to continue in Academia and decide whether you are ready to do that.

→ What was the best piece of advice you have received on your way towards a career after Academia?

That is kind of sad to say, but I think I did not get any advice worth mentioning as “good.” I often got bad advice. One reason, perhaps, is that I have not had the chance to get a mentor so far.

There is one thing I would like to share: As I announced at the university that I was leaving, I had a chat with the person who was responsible for the teaching activities at the faculty, and she told me something I will remember for a long time: “Mr. Fouché, it is great that you are going to industry. It will be a precious asset to have some years of experience in the industry when later you apply for professorships”. This statement surprised me because leaving Academia at first is not commonly seen in people who have become professors (except maybe in applied universities). However, I can see very well how industry experience can be beneficial for leading a chair or a research lab. Ironially, the door I thought I was closing seems to open itself a little more.

I find that the bottom line is that there are two ways to approach a career in general (and it does not matter much, Academia or not):

There is the typical path:

Do your Ph.D. with a famous Professor.
Do a Postdoc in the USA.
Write one or several nature papers.

And, there is another path, where you do whatever you want, stay true to yourself, and things turn out fine because people value your unique experience and want to hire you because you can bring something different. But remember that those positions are rare because people tend to hire people who are alike.

…

If you are reading this, you might have considered leaving Academia. Do you see it differently? And, if you did leave, did you have a different experience? I would be happy if you could share in the comments :).

Would you like to read more about the topic of leaving Academia? Here are a few blogs I found insightful:

On leaving academia, Hadas Kotek
Goodbye, Academia!
Goodbye academia, I get a life.
The Struggling Scientists (If you are more into podcasts)

Almost 30: I am, I have, I can, I want

Wed, 20 Jul 2022 23:00:00 +0000

I am almost 30…

I am much happier than when I was almost 20
I am more open-minded, tolerant, aware of my body
I am an introvert
I am courageous
I am an architect personality type (INTJ-T)
I am often asked for advice
I am happy to talk to those who listen
I am rarely losing my temper
I am a good leader and teacher
I am learning new things every day
I am not believing in any superior entity
I am quite concerned about the world
I am not concerned about others’ opinions
I am sorry for the people I have hurt
I am married to someone I truly love
I am aware I am lucky

I have two master’s degrees, a Ph.D. and a MBA
I have undertaken some challenges
I have failed several times
I have made a few gambles
I have no unhealthy addiction
I have high expectations
I have received several distinctions
I have forgiven my parents
I have difficulties to cope with injustice
I have not given up on humanity
I have no TV, and a relatively old smartphone
I have limited my social media exposure
I have red several personal development essays
I have learned how to use my energy effectively
I have changed political opinions
I have got a few intimate friends
I have made a few foes
I have never been grieving
I have had a few memorable trips
I have moved to Germany permanently
I have realised the importance of family
I have become a scientist
I have not found my dream job yet

I can take pictures; some of them are beautiful
I can deal with computers and technology in general
I can handle frustration
I can make sacrifice
I can leave my comfort zone
I can be on my own at a restaurant
I can have a little humour
I can be quiet
I can listen
I can remain focused over long periods
I can fluently speak three languages
I can explain things clearly
I can care about details
I can deliver
I can cope with uncertainties
I can make good decisions
I can do what no-one expects
I can write this

And now what?

I want to have more time for hobbies
I want to go to museums, theatres, operas
I want to be an artist
I want to become a father
I want to suck the marrow out of life
I want to be recognized for what I do
I want to have an impact
I want to write a book
I want to live somewhere else
I want to solve bigger problems
I want to trust myself a little more
I want to see what I will do next

See you in 10 years?

A few things I learned about doing a PhD in Germany

Mon, 13 Sep 2021 07:00:00 +0000

If you want to be recognized as a scientist, you need to do a PhD. While it is an incredibly fulfilling and transcending experience, working towards a PhD also is a long and tedious journey. There also are many misconceptions about it. If you are interested in research or considering doing a PhD, you may find this article helpful.

Small disclaimer: As I did most of my studies in Germany and in the field of Computer Science, what I am about to say may not apply well in other places or fields. However, I expect most points to be similar, with a few variations in how things are done. If you have had a different experience or opinion, please feel free to share in the comments!

Frankly, when I started my PhD, I did not know very well what I was about to do. I was also very unsure I could actually make it. Below is what I would say to my younger self considering whether to do a PhD or not.

What a PhD degree really is.

Simply put, a PhD is the highest academic degree one can get. Later on, there are further qualifications, like the habilitation or a professorship, but they are no degrees. A PhD says that you have made a significant contribution to advancing human knowledge in a specific field. Getting a PhD is like “getting your wings” as a scientist. It says that you are ready to make sacrifices and that you became an expert at learning. It says that you are an explorer: you enjoy challenges, you can find your way in uncharted territory, you can deal with loneliness, and finally, you know how to tell the world about the fantastic things you have discovered. If you decide to stay in academia, the work done in your PhD may follow you for a while. People (professors, funding agencies) will expect you to pursue related yet different research goals. Even if you do not stay in academia, people will still ask you frequently: “And so, what was your PhD about?”

The PhD degree often is underrated.

People often forget that a PhD is very different from other degrees. There are no teachers, no courses, no exams, no deadlines. You are responsible for setting your own deadlines and find a way to reach them – You are your own boss. I have seen people reacting differently about own-set deadlines at this stage, they either get overly stressed about it, or simply fail to meet them. The truth is that nothing bad happens when you miss a deadline, but nothing good either. You do not progress (which, in turn, is bad). You may have an advisor (sometimes two) and fellow PhD students, but they often are busy and cannot do much work for you; they may only advise you.

Depending on your level of abstraction, i.e., whether you have been focusing on theoretical or applied aspects, companies often try to see you as an inexperienced student, a nerd, sometimes worse: an academic snob. This may be a trick to give you a lower pay by letting you believe that you are not that valuable. The truth is that PhDs are scarce resources. You are not a student; you are a scholar. You have become a master at yourself and know how to conduct (international) projects involving various actors: industry partners, students, advisors, administrative staff. You can handle teaching, staffing, financial and legal aspects while being an expert in a competitive field. The work done during the PhD is a relevant experience for positions in the industry and may help to stand out from the crowd.

The PhD degree is a title.

The pride to be a doctor in Germany goes as far as writing the PhD title on your doorbell, on your ID, on your train ticket reservations, etc.… which I find crazy. This myth of the social status of doctors is very different from my country of origin, France, where almost no one cares. I do believe, however, that one should use the title whenever appropriate. After all, you have worked hard for it. The doctor title is a prerequisite for a career in academia, and it may also open doors later in top-management positions in the industry. In any case, becoming a doctor requires much more than the desire to earn more money in the future – such motivation will not be enough.

Doing a PhD in Europe or the US?

In Europe, PhD and Master’s degrees usually are separated, and a Master’s degree is a pre-requisite for getting a PhD student position. On the other hand, one does not usually have to pass any PhD qualification exams, as typically done in the US. Thus, there is no distinction between “PhD candidates” and “PhD students”. You are evaluated mainly based on your thesis and your defense and (indirectly) the quality of the venues you may have published in. Since Master’s and PhD programs are often bundled together in the US, the typical duration to complete the degree is 5-6 years, while in Europe, it is 3-4 years. In the US, when you “only” have a Master’s degree, people tend to think that you are a “PhD dropout” (i.e., you did not complete the PhD). In Europe, it is the norm not to pursue a PhD after the Master. On the other hand, we often think that the Bachelor (usually a three-year curriculum) is a relatively short academic education.

Doing a PhD in Germany.

I must say that I have been fortunate with my own PhD experience. I did my PhD in Computer Science at the Karlsruhe Institute of Technology (known as “KIT”) – a top university in Germany. Germany has a long-standing academic reputation, and PhD degrees are very well recognized in German society.

In my field, there are two types of PhD titles, the Dr. rer. nat. (Doktor der Naturwissenschaften – Doctor of Natural Sciences), and the Dr.-Ing. (Doktor der Ingenieurwissenschaften – Doctor of Engineering). While the latter is sometimes considered more applied than the former, they both have the same value. Academia tend to prefer Dr. rer. nat. while industry tend to prefer Dr.-Ing. I am personally a Dr.-Ing., but I have met numerous Dr. rer. nat. doing much more applied work than me. I find the Dr.-Ing. appropriate in my case because I do see Computer Science as an engineering discipline. Traditionally, one tends to inherit the same title as their advisor. In principle, the doctoral committee (composed of your advisor and further examiners) decides which title you get.

In Germany, the conditions to do a PhD are excellent. As a PhD student, you get a salary based on a civil servant pay scale (TVL-E13, you may look it up here), which is more than enough to cover your living expenses. In my case, this was only about 15-20% less than the salary I would have got by starting in the german industry right after graduation as a software engineer.

Your status is regulated with a formal employment contract at the university that lets you save for your future retirement, with vacations days (30 per year!), working time (40 hours per week), and unemployment benefits once the contract expires. You are not a student anymore. However, doing a PhD often takes much more than 40 hours per week, so do not count too much on that!

Depending on your field, you might not get a 100% position, but those are very common in Computer Science. In comparison, one does not get such good conditions in France or many other countries. In France, the only way to get similar conditions (that I am aware of) is to do your PhD in the industry, which often comes with additional duties besides the actual PhD work.

Doing a PhD at the university or in the industry?

Traditionally, PhDs are done at the university. You are integrated into a chair and expected to engage in teaching activities (taking care of exercises, correcting exams, giving lectures sometimes). Depending on the level of involvement in teaching, the PhD may last up to 6 years, but that depends very much on your advisor and your progress in your research. I did my PhD at the university within a little more than three years.

There are also ways to do a PhD in the industry. In France, this is best known as a CIFRE thesis. It can be a way to earn more money while doing your thesis (but that is not always the case). In Germany, major companies tend to offer a few PhD student positions. They are special contracts in which you are expected to work on a research topic deemed critical to the company, while you typically also work on other projects (often 50/50). A PhD in the industry typically is more applied and is often challenging due to the additional tasks, and they often are strictly limited to 3 years.

How to find a PhD student position.

This is, in my opinion, the most challenging part. Except for a few specific PhD programs and fellowships, PhD student positions are not advertised online. It is easy to understand why: The professors and postdocs of the group have to handle the recruitment process themself, among many other tasks. They are likely to get hundreds of applications to deal with by advertising the position, which lowers their available time for research and teaching. Thus, if the position is ever published, it will be on the website of the corresponding group, and it is often hard to find (on purpose).

Positions are typically given to outstanding students, i.e., they had excellent grades or did impressive work in their bachelor/master thesis. In that case, they may get directly contacted by a potential advisor (that was my case). If you are actively looking for a position, you may ask a professor who knows you if they have an open position or recommend you to another professor (but that only works if they know each other somehow). If you are actively looking, you must already have an idea about your PhD thesis project, i.e., known some of your research interests. Read recent papers of the professor(s) you want to work with and make a few meaningful proposals for future work. This way, you may catch their attention. Be proactive!

In my opinion, a great thesis is the best indicator for an advisor (and for the student!) that one can conduct a good PhD project. In my field, the expectations from student theses are similar to those from a PhD student for a short period of time, so one can see the student thesis as a trial period for a PhD. Often, the Master’s thesis serves as a basis for your first PhD paper. Since you have done it before your time as a PhD student, it may even give you a head start! Keep in mind that professors/postdocs are almost always looking for talented students; the thesis is your opportunity to show your potential. If you are applying for an advertised position, you must prepare a strong application, as the competition is fierce.

How to fund a PhD student position.

The funding of your position usually comes either from: (1) the group’s university allowance for teaching, (2) some large grant like a graduate school or a so-called research training group, or (3) an individual fellowship for which you would have to apply personally upfront. In my case, most of my funding came from my participation in a research training group on energy data (so, (2)), and indirectly from being a member in the Software Campus program (which is like (3) + some industry collaboration).

Depending on the funding type, the overall experience tends to be slightly different. (1) involves more teaching. (2) involves participating in interdisciplinary research or large projects. (3) tends to give you more freedom and often comes with networking opportunities with other fellows. Typical funding schemes tend to vary widely from one field to another.

Two PhD models: the monograph vs the cumulative dissertation

There are two PhD models in European universities: (I) the monograph and (II) the cumulative dissertation. In (I), you work hard for several years and publish one large piece of original research, the doctoral thesis. This is the traditional way of getting a PhD, but it is now somewhat outdated. In (II), you also work hard for several years, but the final dissertation is more or less the concatenation of several units that you have published over those years.

I believe that (II) is a natural response to the ever-evolving pace of research. If you were to publish five years of research all at once, the chances are high that similar studies have been published meanwhile, making your contribution already outdated. Especially in my field (Data Science), the state of the art tends to change every few months. Another reason is that (II) makes it easier for your doctoral committee to evaluate the quality of the work: If you published in top conferences, then your work must be of top quality, right? Sadly, that is not always the case, but the quality of the venues is often used as a proxy for evaluating the quality of your work.

In my case, I did the cumulative model (II), but I also had to bundle the individual contributions together as a monograph (the doctoral thesis, so actually (I)). In that case, the individual contributions must relate to each other to some extent because you need to tell a bigger story. You must plan strategically.

Two PhD starts: the defined topic vs. the open topic.

There are two main ways to start a PhD: (A) you get a position where the task(s) to fulfill are known to some extent and have been worked out for you, or at least some of the research questions have already been identified by a senior researcher. The second way, (B), lets you start without any clear goal, except getting a PhD by contributing a subfield (e.g., Data Mining) of your discipline (e.g., Computer Science). Both ways (A) and (B) have their pros and cons.

(A) – the well-defined topic – has the following features:

You are likely to address a topic that fits nicely in a larger scheme or project.
Thus, there will possibly be more synergy with other members of the research group.
The topic is likely not too small, not too big, so the scope is adequate for a PhD student.
Initially, this was not your idea, so it might be harder to fully embody it, live it, and defend it.
Your advisor(s) might have specific expectations on addressing the topic, limiting your creativity.
At the start, it is unlikely that you have expertise in the topic, so you are facing a steep learning curve.
You may end up feeling like a cog in the wheel of someone else’s academic career.

On the other hand, the open topic (B) is as follows:

You have total freedom to explore whatever YOU want.
You may get a head start by focussing on an area where you already have some knowledge.
No one can tell you what to do; you are expected to master this topic on your own.
You obtain more distant supervision. You must develop a high tolerance for uncertainty.
Consequently, you are likely to feel lonely or isolated – and to have motivation problems.
Aim too small, and people may not be very impressed. Aim too high, and you will not finish.
You are more prone to imposter syndrome and burnouts.

Of course, everything is not black or white; you may get something in-between (parts of the research are clear, but you must solve new questions on the way). In my case, I was dealing with a type (B): a very open (or I would say, even undefined) topic. Which type you get changes your approach and your experience as well. I think that type (B) is more common in Germany and university, while type (A) is typical in France and the industry.

The duration of a PhD degree.

I was in a group where students are expected to graduate within three years. In Germany, it is common to take up to 6 years (sometimes even more). The main reason is that you often get additional teaching duties besides your PhD. Depending on your teaching workload, you may not have much time for research. If you are good at teaching (and if you enjoy it), your advisor might want to keep you around for a few more years as well by postponing your graduation. You may not receive money from the federal government for your PhD for more than six years by law. If you have third-party funding, this does not apply. This is the same for the time as Postdoc.

After three years I was not done because I needed almost a year to define my topic (a typical type (B) start problem). I finished my PhD in 3.5 years, and I was fortunate to get a contract extension to complete the thesis without experiencing the stress of being unemployed. It is more common in many places outside of Germany to only have three years and nothing more (even if you are not done).

Is it possible to fail one’s PhD degree?

Not really. What happens is that time may run out, and you did not manage to produce enough evidence to defend your thesis convincingly. Except for a few sporadic cases, one does not fail in the PhD defense. Your advisor will stop you from defending your thesis if they feel that you did not contribute enough to submit your thesis. Some students finish their thesis under unemployment or as a side activity besides a new job when time runs out (this is very difficult, try to avoid this). Some students lose motivation and never finish. Some students find out early as that they actually do not want to do a PhD, and that is ok.

Although the perspective of not bringing a PhD to completion, i.e., failing, might sound like a bitter outcome, one must also realize that it is just normal. After all, it is just very hard, and one needs some luck. If the sacrifices required to get a PhD exceed what one is ready to make, it is ok to walk away and move on with life. In the end, not finishing a PhD only is a failure if you see it as such. Also, even as a PhD dropout, job prospects usually are excellent in the industry.

Concluding remarks

So in case you are in the process of thinking about whether to do a PhD or not, I hope that you find the information here helpful. In my opinion, it is a very personal choice. I can nonetheless highly recommend it – This is a unique opportunity to learn about your field, the process of science, and yourself. Do you have a different opinion? Does it work differently in your field / geographic area? I would be curious to hear about it! Next up, I will try to compile a set of specific pieces of advice for people deciding to go for it.

Simple Workout Exercises for Confinement

Mon, 01 Feb 2021 07:00:00 +0000

Exercising is extremely important for our physical and mental health. Nowadays, it could not be more true, as we are facing numerous restrictions: We can’t go to the gym, we can’t meet our friends and family, we work from home and don’t have much time outdoor. Also, many of us live in tiny spaces. In this article, I am sharing a small set of exercises that are suitable for our confined lives. Maybe you’ll find them useful too!

How I came up with it

Let me start with some background. In Q4 2019, I visited UIUC in Champaign (IL, US) for 3 months for a research project. I rented a small room in a house¹ and made very good friends at the local Roller Derby² leagues, the Twin City Derby Girls and the Prairieland Punishers. I wanted to experience Roller Derby in the US but, at the same time, I was afraid to hurt myself, and I did not want to use a foreign health system (and we all know that in the US it is crazy expensive…). So I came up with this small set of exercises that I would do almost every day in the evening. I bought for $20 a small graphic tablet to a student and drew it to increase my commitment and share it with peers!

DISCLAIMER: I am not a health or sport expert. There are many great resources out there. A well-known reference for confined exercises is the book by famous prisoner Charles Bronson: Solitary Fitness. But hopefully, this one-pager is a good start for someone to exercise on their own or with friends.

The Daily Derby Hell Workout

It appears that this browser does not support embedded PDFs. Please download the PDF to view it: Download PDF.

Put the exercises together, and you get a nice routine, which I call the Daily Derby Hell Workout!

By the way, Dr. Frog is my Derby name ☺.

For Sally, here is the song: Flower (Moby). At “Bring Sally up”, you go up. At “Bring Sally down”, you go down.

Advantages

The advantages of this routine are as follows:

You need almost nothing!

I recommend having 2 Gallons of water (as weights), perhaps one more for drinking, and a stopwatch or phone. It is nice to have some music, a mirror (to check your posture) and a balance pad (but you can fold a blanket 4 times for a similar effect). It is easy to do this routine when you travel or don’t have the equipment.

You work on your core muscles!

So you reduce your risk of getting injured and improve your posture. This is also good for practicing ankle and back intensive sports like Roller Derby.

You can do it almost anywhere, any time!

You can adjust the time and difficulty on each exercise depending on your needs and level. I recommend you to spend up to one hour on this routine. Switch the order of the exercises as you see fit and to avoid getting bored. The exercises do not require you to jump around so you can do them in tiny spaces as well.

You can share it!

You can call your family or friends via Skype, Zoom, or any videoconferencing tool and do the routine together! The PDF is a creative common material (CC-BY-SA-4.0) in DIN A4 format.

Results

I followed this routine for about 3 months and ate as healthy as I could (organic vegetables, as much as possible) and gained about 4 kilos on top of my modest 60 kilos, no fat. This routine did not only help me preventing injuries (I did several scrimmages, had Roller Derby 2-3 times a week and went to the skate park on the weekend), but also helped me to be successful in my research stay. After that, I could also see a notable difference in my performance as a skater and generally felt stronger and more balanced in my body.

You can enhance the workout further by adding other exercises (there is room for notes in the bottom-right corner) and embed it in a Tabata routine. If you are not familiar with Tabata-style workouts, the concept is simple: You do W series (typically 2 to 3) of X exercises (typically 10 or more), each for Y seconds (typically 60) with Z seconds of break in-between (typically 20). There are great free apps out there to help you manage time, like this one. In the future, I will be sharing some of our Tabata series (I say our because we are putting them together with Charlotte, my significant other).

I think the routine would benefit from more exercises for legs. Do you see further exercises to include in the routine? Please share them via the comment section!

Take care everyone and stay healthy!

Actually, not any house in Champaign, the Ivy House! It is an art gallery with a few rooms for rent on the second floor, which I found via Airbnb: https://www.instagram.com/ivyhousegallery/ ↩
In case you don’t know it, Roller Derby is a roller-skating contact sport played by two teams of fifteen members. See this one-minute introduction: https://youtu.be/OId6gTd2LCM ↩

arXiv and the Information Explosion

Wed, 18 Nov 2020 07:00:00 +0000

Nowadays, it has become more and more common to publish on arXiv before peer-review. There are several (good) reasons for that: (1) Academic publishing takes too much time, and even then, (2) your paper often ends up behind a paywall. In over-hyped fields, the state of the art may change every month (or sometimes, week!), so it is a good way to “plant the flag” on your research idea. But what are the consequences of that?

This indeed comes with some drawbacks. Before it was “cool” to publish on arXiv, researchers had no other option but to publish in peer-reviewed journals or conferences, which only have room for a few hundreds of papers per year. As of 2020, we are now facing more than 44K paper per year being published on arXiv, just for the field of Computer Science! This more than 155K per year, all categories together.

Even if scientists tend to specialize in small niches, it becomes challenging to stay up to date with the progress in your field. If you were to read, say 1K new paper per year, that would be about 3 per day – and you also need to allocate time to your research. As a researcher, it just becomes harder and harder to decide what to read. So, will the number of submissions per year continue to increase? Let’s find out.

Mining the arXiv Metadata

You can reproduce all the plots from this GitHub repository. Note that arXiv does publish some statistics. For example, here. However, the plots are somewhat clunky. We obtained the MetaData about every paper submitted on arXiv from here.

Warmer colours are for categories with more submissions (in 2019). As we can see, Computer Science (cs) only recently become a major field on arXiv, but it is increasing fast! Let’s have a look at the relative counts:

So the category hep-* (High Energy Physics), has been most prominent for a long time, now it is math and physics. Let’s see how fast the total number of paper (the size of the arXiv!) has evolved over time:

As we can see, the number of articles has doubled twice in 1992, then doubled every 1-2 years, and now every 5-6 years. Can we guess how many papers will be on arXiv in 2030? Let’s plot the evolution of the number of submission per year.

We have also fitted a second- and third-order polynomial to the evolution of the number of submissions. The third-order polynomial fits the bars quite nicely but leads to a steep increase. In comparison, the second-order polynomial feels more realistic. With the second-order polynomial, we forecast that there will be about 3.6M papers on arXiv in 2030, while the third-order polynomial expects about 4.1M.

Who is right? Only time will tell. It depends on how many scientists in the world are there, who do not yet use arXiv —Rendez-vous in 2030.

And what about Computer Science?

As we could see, the number of submissions has increased steeply in recent years. Let’s have a look!

That is a much faster increase as before! What about the relative counts?

So at the very start, there were only three categories: DS (Data Structures and Algorithms), CC (Computational Complexity), and GL (General Literature). In the early 90s, the AI (Artificial Intelligence) community massively submitted to arXiv, but the larger fields are now CV (Computer Vision), IT (Information Theory), LG (Machine Learning), and CL (Computation and Language).

As an exercise for you: How many computer science articles will we have on arXiv in 2030? Of course, you can use the polynomial-fitting technique above, but the question is instead: How many more paper per year will the community post on arXiv?

Threats and Opportunities

As we have more and more papers, each of them gets less and less exposure (unless if you work with someone famous?). In the end, one may only consider those with more citations. But getting citations is a tricky business (more on that later, probably). Although citations are often being used as a proxy for quality, they do not help us to find new problems, i.e., those which did not get much attention so far!

In the future, it may become more likely that we will reinvent things over and over again, simply because they have gone unnoticed. By systemically exploring the papers being published (via Data Mining, Machine Learning, etc.), can we find new trends? New subfields?

The massive information explosion on arXiv opens the ways to exciting research directions. arXiv can be a great tool to explore the historical evolution of a particular field. For this, however, you might need to dive deeper into the semantics of the papers. Luckily we now have many great tools to extract semantic information from text! Interesting questions are, for example:

What has been the focus of community X over the past few years? How did it change?
What are the common interests between community X and Y? How much do they overlap?
Can one see community X and a parent of community Y?

Getting answers to such questions may be not only interesting from a historical perspective but also to identify new trends and emerging fields! Finally, it would be nice to visualize the relationship between communities at different points in time and create better taxonomies.