How to actually learn from a video

A while back I watched an eleven-hour course on distributed systems. Two speed, mostly, over about a week. I finished it on a Thursday. The following Monday someone asked me to explain quorum reads and I produced roughly one sentence, most of which was the word "quorum."

That was annoying enough that I went looking for the reason, and the reason turned out to be boring: I had spent eleven hours watching someone else be competent. That is a genuinely pleasant activity. It is also almost completely unrelated to becoming competent yourself.

What fixed it wasn't a better course. It was a change in what I thought the video was.

Video pretends to be a lesson. Treat it as raw material.

A lecture in a room has friction built in. The speaker goes at their pace, you can't rewind, and if you drift for thirty seconds that material is gone. That friction is annoying but it forces you to keep up in real time.

Video removes all of it, and then removes a bit more. It plays whether or not you are following. There's no moment where the thing stops and looks at you. You can watch a two-hour lecture in a state that is functionally identical to watching a two-hour movie, and at the end you'll have the same feeling of completion — which is the trap, because the feeling of completion is doing no work at all.

So: the video is not the lesson. The video is a pile of material that you turn into a lesson. The turning is the part where the learning happens, and it is not passive, and it is slower than watching. That's the whole trade.

Rule one: your hands do something every few minutes

The single highest-leverage change is to make sure you are producing something, continuously, while the video plays. Not at the end. During.

Notes are the obvious version of this, and most people's notes are bad in a specific way: they're transcription. If your note says "consistent hashing distributes keys across nodes using a ring," you've moved words from the video into a document without them passing through you. That note has the texture of understanding and none of the content. You'll reread it in a month and it'll be a stranger's handwriting.

The test I use is: could I have written this note before watching? If the sentence is generic enough that it could open a Wikipedia article, it's transcription. Rewrite it until it contains something only you would write — the confusion you had, the example you thought of, the thing you thought it meant before it turned out to mean something else.

Some things that pass the test:

• "This is basically the same idea as the load balancer thing at work, except the failure mode is different because —"
• "Wait, why doesn't this break when a node comes back?"
• "He said 'obviously' here and it wasn't obvious. Come back to this."
• "Three things a quorum read guarantees, in my words: ..."

That last one is the workhorse. Say it in your own words, badly, without pausing to make it elegant. Badly-in-your-own-words beats well-copied every time. The awkwardness is a sign the sentence went through your head instead of around it.

Rule two: pause on the fork, not on the summary

People pause too much and in the wrong places. Pausing every forty seconds turns a lecture into an hour of stop-start and mostly trains you to be irritated.

The moments actually worth stopping for have a shape. They are the moments where the speaker is about to tell you something you could have predicted, and haven't tried to. A worked example before the answer. A question they pose rhetorically. The instant after they set up a problem and before they solve it.

Stop there and guess. Out loud, in a note, doesn't matter. You'll usually be wrong, and being wrong for four seconds before hearing the right answer sticks harder than hearing the right answer cold. This is the closest thing to a free lunch in the whole business — it costs almost nothing and works even when you're tired.

Do not pause to write a summary of what was just said. That's transcription with extra steps.

Rule three: 1.5× is fine, 2× is usually a lie

I'm not going to tell anyone to watch at 1×. Most lecturers speak slowly, and the first pass over familiar material genuinely can be sped up.

But speed interacts badly with difficulty in a way that's easy to miss, because the thing speed destroys first is your ability to notice you're lost. At 2×, keeping up with the audio consumes the attention you'd otherwise use to think "hold on, I don't follow." You stay in a comfortable state of near-comprehension for a long time, and near-comprehension is indistinguishable from comprehension from the inside.

So a practical rule: fast for the parts you could roughly explain already, and drop to normal speed the moment something new lands. If you can't remember the last time you changed the speed, you're probably watching the hard parts at the same rate as the easy ones, and paying for it later.

Rule four: timestamps, so future-you can get back

Notes taken from video have a failure mode that notes from books don't. A book note can be wrong or thin and you can find the passage again in a minute. A video note that says "he explains the retry thing well" is, three weeks later, an eleven-hour haystack.

Timestamp anything you might want to return to — a proof you half-followed, a diagram you don't want to redraw, a claim you're not sure you believe. The point isn't to catalogue the video. It's that "I'll come back to this" only means something if coming back is a click.

This turns out to matter more than it sounds like, because it changes what you're willing to skip. When returning is cheap, you can let a confusing five minutes go, mark it, and keep momentum — instead of grinding on it, losing the thread of the argument, and quitting.

(This is the one part of the whole process that's genuinely tedious to do by hand, which is a large part of why Floweo exists. Ctrl then t in the editor drops the current playback position into your notes as a link. But a text file and manually typed 14:32 works fine — the habit is what matters, not the tool.)

Rule five: close the video and write the thing you can't

Here's the step almost nobody does, and it's the one that decides whether any of this stuck.

When the video ends, close it. Then, from memory, write down what it was about. Not the outline — the argument. What problem was being solved, what the solution was, why the obvious simpler approach doesn't work.

You will immediately hit a wall. There'll be some link in the chain you cannot reconstruct, some step where you know that it works but not why. That wall is the useful output of the entire session. It's the only reliable signal about what you actually don't know, and it's completely invisible while the video is playing, because while the video is playing the speaker's confidence is doing the work for you.

Then go back — via your timestamps — and watch only those bits. Five minutes of rewatching aimed at a known gap is worth more than a second full pass, by a lot.

Rule six: one week later, ten minutes

Nothing survives without a second contact. Everyone knows this and almost nobody does it, because the first contact felt so productive.

It doesn't need to be a system. A week after the video, look at your notes for ten minutes. Specifically: read your questions, not your answers. Can you answer them now? Cover the note and try to say the idea before you read it. Anything you can reproduce cleanly, you can stop worrying about. Anything that's gone fuzzy is where the next ten minutes go.

Ten minutes, once, a week later, gets you most of the benefit of any spaced repetition scheme. Whether you graduate to a real system later is a detail.

The uncomfortable part

All of this makes video slower. An hour of lecture becomes ninety minutes or two hours of work. That feels like a downgrade, and for a while it's genuinely less pleasant — you lose the pleasant flow of watching a smart person explain something well.

But the eleven-hour course I couldn't explain cost eleven hours and returned close to zero. The alternative is that a two-hour lecture costs four hours and you keep it. It isn't slower. It's the first version that has a rate at all.

The frustrating truth about learning from video is that the medium is optimised for feeling like it's working. Everything above is really just a set of tricks for interrupting that feeling often enough to find out whether it is.