I have seven notebooks on my shelf. Not a curated number — just however many accumulated across my student years at Waseda and the years of professional work that followed. Each one is a kind of sediment layer: I can pick up the third one from the left and open it to a random page and know immediately whether I was alone when I wrote it or whether I had just left a group session. The handwriting is different. The structure is different. The quality of thinking is different, too, though not always in the direction you’d expect.
I’ve been thinking about this split — solo versus group — for a long time, and I’ve arrived at some conclusions that contradict the standard advice students get. The standard advice is to find a study group, to show up, to collaborate. That advice is not wrong exactly, but it is incomplete in ways that cost people real hours. I wasted those hours myself before I understood what I was actually doing.
The Waseda algorithms group
In my second year at Waseda, I was taking an algorithms course that I found genuinely difficult — not conceptually impenetrable, but the kind of difficult where you understand a thing in lecture and then sit down to implement it and realize the understanding was shallow. I joined a study group that met twice a week in the library, five of us, all from the same seminar. We would spread our textbooks and laptops across one of the long tables near the windows on the fourth floor and work through problem sets together.
For about six weeks I thought this was going well. And in certain ways it was. When I explained something — when I stood up and drew a binary tree on a piece of paper and walked the others through a traversal — I understood that thing afterward in a way I had not before. There is something real about the pressure of explanation. You cannot explain what you do not know, and the moment someone asks a question you cannot answer, you feel exactly where the gap is. I learned the conceptual architecture of algorithms faster from that group than I would have alone. I am reasonably certain of that.
But I started keeping a log of each session. Not formally — just a line at the top of my notes: arrived time, left time, topics covered, time lost. That last column was the one that changed how I thought about the group. Forty minutes per session, on average, was what I calculated as social drift. Not a ten-minute break — I am fine with breaks. I mean the kind of drift where someone mentions a movie and that becomes a fifteen-minute conversation, or where two members of the group get into a prolonged side debate about something peripheral, or where we spend twenty minutes deciding what to work on instead of working. Forty minutes, twice a week, for six weeks: I had given that group roughly eight hours of time that produced nothing academically useful.
I did not quit the group angrily. I just stopped scheduling around it. I told the others I would come when I was stuck on something conceptual, which was honest, and I did show up for two more sessions when I genuinely needed to think through a problem out loud. But I stopped treating it as a regular obligation. My studying became sharper after that, and my notes from that period — the third notebook on my shelf — reflect it. The handwriting gets more regular. The Cornell-method columns I was using at the time get denser in the right margins. I was spending that reclaimed forty minutes on actual recall practice.
What I learned about when groups work
What I took from the Waseda experience was a distinction I now apply to every study or work situation I find myself in. There is a category of cognitive task where other minds genuinely help — where the presence of another person who knows things you do not, or who will ask questions you would not ask yourself, or who needs you to explain something and thereby forces you to confront your own gaps, accelerates learning in a way that solo work cannot replicate. And there is a different category where another person is mostly friction: they interrupt the pace you need, they pull your attention toward their level of understanding rather than yours, they transform a focused hour into a negotiated hour.
Conceptual material — the kind where you need to understand the why, not just the what — belongs mostly in the first category. When I was trying to understand why certain graph algorithms have the complexity they have, talking through it with someone else was genuinely useful. My understanding after the conversation was different from my understanding before it, not just reinforced but restructured. I cannot achieve that restructuring alone, or at least not as efficiently. The resistance of another person’s comprehension, their questions, their wrong answers that I have to correct — these are not noise, they are signal.
But memorization belongs firmly in the second category. Reviewing flashcards with another person is almost always slower and stickier than doing it alone. The rhythm of Anki or any spaced-repetition system is a private rhythm — it adapts to your errors, not to some average of your errors and your partner’s errors. I have never met anyone who convinced me that group Anki was a good use of time. Writing belongs in the second category too. I am a programmer now and I write a lot of documentation and the occasional technical post, and I have never once produced better prose with someone watching. The writing that comes from a group is committee prose, and committee prose is readable in the way that beige walls are readable: it communicates and that is all.
Practice problems are the interesting middle case. An easy problem that I know how to solve, I should just solve it. A hard problem that has me stuck — genuinely stuck, not just uncomfortable — is where another person can be worth more than an additional hour of solitary frustration. The important qualifier is genuinely stuck. I have a rule I apply now, as a remote programmer: if I have been stuck on something for more than twenty minutes and my last three approaches have been variations on the same wrong idea, I message a colleague. Not a scheduled session, not a meeting — just a message. Hey, I am looking at this for twenty minutes and I keep doing the same wrong thing, do you have five minutes? That is all I need. The insight usually comes in the first exchange.
Parallel solo: the third option I did not know I needed
When I moved out of student life and into remote work, I thought I had left the question of solo versus group behind. But working from an apartment in Setagaya, some days entirely alone, I ran into a different version of the same problem. Pure solo work for eight or nine hours produces a kind of mental staleness in me that I notice around the third or fourth hour. It is not that I cannot concentrate — I can still concentrate, but the concentration has a slightly mechanical quality, like I am executing familiar procedures rather than thinking.
My colleague Yuta and I started doing what I now call parallel solo sessions. We open a video call, cameras on, and we work on completely separate things. He might be debugging something in Rust while I am writing a design document. We do not discuss our work. We do not check in. We are just present on each other’s screens, doing our own work, with the ambient awareness that another person is also working. We end the call after ninety minutes or two hours and go back to our separate tasks.
This sounds like it should not work, and I cannot entirely explain why it does. My best hypothesis is that the visible presence of someone else working removes the low-level social isolation that quietly degrades my concentration during long solo sessions. I am not lonely when Yuta is on the screen, even though we are not talking. The slight social warmth of presence is enough. Crucially, I do not lose the advantages of solo work — I set my own pace, I do not have to explain my thinking, I do not get pulled into his problems or he into mine. Researchers call it the body-doubling effect, usually discussed in the context of attention difficulties, but my experience suggests it applies more broadly.
Parallel solo is now my default for long work sessions. Pure solo for shorter, intensive bursts that require me to hold a complex structure in working memory for an extended stretch — the first two hours of a hard writing session, typically. Parallel solo for the middle portion, when I have the structure built and I am populating it. Message Yuta when I am stuck for more than twenty minutes and my thinking has stopped moving.
Why most study groups fail
The fundamental problem with most study groups is that they are organized around presence rather than purpose. We will meet Tuesday and Thursday, the implicit logic goes, and something educational will happen because we are all there. But presence is not a learning mechanism. It is just presence. Without a specific problem to solve, a specific concept to teach, a specific piece of material to work through collectively, a group session defaults to the path of least resistance — which is conversation about something other than the work.
I understand why the Waseda group drifted. We were tired students, the work was hard, and talking was easier. The drift was not a character failing — it was a predictable response to an environment with no structure to resist it. If I could advise my second-year self, the advice would be simple: every session needs a written agenda before anyone shows up. Not “review Chapter 5” — something specific: three problems we will attempt, one person teaching concept A, another teaching concept B, done in ninety minutes. That kind of session can work. The open-ended “let’s study together” almost never does.
Preparation is the other essential. A group session where one person has not done the reading is not a study group — it is a tutoring session for that person, paid for by everyone else’s time. When all members arrive prepared, the group can do higher-order work: challenge each other, find gaps in each other’s understanding, argue about edge cases. When even one person arrives unprepared, the session regresses to baseline coverage, and the prepared members gain nothing they could not have gotten alone.
The size question is also not trivial. My experience is that two or three people is the functional range for most academic work, and four is about the maximum before coordination overhead starts eating into productivity. Beyond four, managing turn-taking and sidebar conversations takes more attention than I want to give it. The discussions become less precise. Large groups also tend to develop informal hierarchies — one or two confident people do most of the talking, the quieter members follow along passively, and the passive members learn less than they would studying alone because passive attendance is not active recall.
A quiet study spot — the kind of environment that supports either solo or paired study
The environment question
There is a version of “studying in a group” that is really just socializing with books open, and I think many students know this about their own sessions but find it easier not to examine. I have been in those sessions. They feel like productivity because you are in the same room as your textbook for three hours. But being near a textbook is not the same as processing its contents.
The physical environment shapes the session in ways that override intention. Someone’s living room with a game console visible is not where serious work happens. A library study room with a door that closes is. A lively café is not where I can do the focused reading that requires holding a long argument in my head. I do most of my reading at home, early morning, before I have spoken to anyone, with a familiar ambient track — usually something from the lofi stream — as a context cue that my brain now associates with a particular mode of attention, not conversation.
When I do work in public spaces now, I choose them deliberately. A coffee shop where I know I will not see anyone I know, where I am comfortable being anonymous for two hours, is a reasonable working environment for me for certain kinds of writing. Not for anything that requires sustained concentration on a problem I do not yet understand — that I do alone, in quiet.
What the research mostly confirms
The academic literature on collaborative versus individual learning is less decisive than education advocates often suggest, but a few findings hold up. Explaining material to others improves the explainer’s understanding — replicated enough times that I treat it as reliable. The mechanism is that explanation forces you to identify and fill your own gaps, which passive review does not. The “Feynman technique” is basically just systematized explanation, and it works for the reason any explanation works: it reveals what you do not actually know.
Peer discussion benefits conceptual understanding more than procedural skill. If you are learning how a data structure works conceptually, talking it through helps. If you are drilling implementation, do it yourself — these are different cognitive tasks and they respond differently to social context. Retrieval practice — testing yourself, not reviewing — is one of the most robust findings in learning research, and it is fundamentally a solo activity. My Anki decks from the Waseda years got me through exams that group review sessions alone would not have.
My current approach, for what it is worth
My current approach to any sustained intellectual work is probably not generalizable in its specifics, but the underlying logic might be. I work alone by default. Solo for all writing, solo for all deep reading, solo for implementation work. I run parallel solo sessions with Yuta for long work stretches where isolation starts to degrade my quality. I message a colleague when I am genuinely stuck — not when I am merely uncomfortable, which is a different state and one that I should sit with rather than escape — but when I have been in the same wrong loop for more than twenty minutes.
What I do not do anymore is schedule regular group sessions as a general productivity strategy. That frame — show up, something will happen — does not match how I actually learn or work. What I do instead is treat group interaction as a targeted intervention: I know approximately what kinds of stuckness are unstuck by another person’s input, and I reach for that intervention when I am in that kind of stuckness. The rest of the time I work alone, or near Yuta in comfortable silence.
The seven notebooks on my shelf contain a lot of evidence about what has worked for me and what has not. The ones from the Waseda algorithms period have denser pages in the sections I worked on alone and sparser, less careful pages in the sections I worked through mainly in group sessions. That is not a random distribution. My handwriting in the group-session notes is looser, the margins less used, the summaries more superficial. I was processing things less fully. The group was doing some of the cognitive work that I should have been doing myself, and that work did not transfer back to me reliably.
The parallel solo sessions with Yuta are genuinely valuable to me, and the occasional twenty-minute call where I explain a stuck problem to a colleague and feel the problem shift — that is valuable too. But the lesson is to be precise about what kind of help you are actually getting from a group, and to compare that honestly against what you would get from the same time spent alone. The question is not whether group study is good or bad. The question is whether this particular session, on this particular material, in this particular configuration, will produce more learning than solo work would. I try to answer that question before I commit to a format, not after I have already spent two hours on drift.
For the work I do now — code, documentation, technical thinking, occasional writing about study habits from a shelf full of old notebooks — that answer is almost always the same. Start alone. Add people when the specific problem asks for them. Know which problem you are actually solving before you decide who should be in the room. A reliable ambient environment supports that solo work more than most students expect; my own routine pairs early-morning quiet with a familiar audio track — usually the 24/7 lofi stream — as a low-friction context cue that this is work time, not conversation time.
