Small ideas, carefully kept.

The result is called inattentional blindness: failing to see something fully visible because attention is engaged elsewhere. It is not a failure of the eyes. The light from the gorilla reached every viewer's retina.

Two details in the study make it more interesting than a party trick ¹:

• Harder tasks deleted more. Viewers keeping separate counts of bounce passes and aerial passes noticed the gorilla less often than those keeping a single count.
• Similarity mattered. Viewers tracking the team in black noticed the black gorilla more often than those tracking the team in white.

So attention behaves less like a spotlight that adds light and more like a filter tuned to a goal. Whatever does not match the goal is treated as noise, even when it is a gorilla.

Ebbinghaus was his own and only subject, which is a reason for caution. But the shape of his curve has been found again and again: steep at first, then shallow.

The practical consequence is spacing. Reviewing just as something begins to slip, and then waiting longer before the next review, makes each review count for more. A large review of the research found that spreading practice out beats cramming it together, and that the best gap grows with how long you want to remember ².

A reasonable starting rhythm: revisit something the next day, then after a few days, then after a couple of weeks.

In the study, one group read a short passage four times. Another read it once and then tried to write down everything they could recall, three times over ¹.

Five minutes later, the rereaders did better. A week later the result reversed: the group that practised recall remembered about 61 percent of the ideas, the rereaders about 40 percent. The rereaders were also more confident they would remember.

That gap between how learning feels and how well it lasts is the useful part. Fluency, the smooth sense of “I know this”, comes cheaply from rereading. Retrieval is effortful, and the effort is what strengthens the memory.

The surprising part is not the size of the budget but how little of it changes. When you start a demanding task, the extra energy the brain uses in the regions involved is small compared with what it was already spending ¹.

In other words, the brain is never idle. Much of its cost is the constant work of keeping neurons ready to signal and of activity that goes on whether or not you are concentrating. A hard problem adds a little on top; it does not switch the engine on.

Exclusive or (XOR) answers “yes” when exactly one of two inputs is on:

• neither input: no
• input A only: yes
• input B only: yes
• both inputs: no

A simple model neuron adds its inputs and fires above a threshold, so “both” always looks like more than “one”. It cannot say no to both. That limit is why XOR became the standard example of what single units cannot do.

The dendritic spikes in the study behaved differently: they were largest near threshold and shrank as input grew stronger ¹. A response that falls off with too much input is exactly what XOR needs. The finding does not make a neuron a computer, but it suggests a single cell is closer to a small network than to a single switch.

Boltzmann’s formula, carved on his tombstone, is S = k log W: entropy S grows with the number of arrangements W ¹.

With four coins the difference is modest: six ways to be half and half, one way to be all heads. With the trillions of trillions of molecules in a cup of coffee, the mixed states outnumber the tidy ones by margins so large that “unlikely” becomes “never seen”.

That is also why entropy is tied to the direction of time. A system wandering among its possible arrangements will almost certainly drift toward the overwhelming majority. The puzzle, as Carroll argues at length, is not why entropy rises but why it started so low ².

The distance changes through the year, because Earth’s orbit is slightly elliptical, so the travel time runs from roughly 8 minutes 10 seconds in early January to about 8 minutes 27 seconds in early July.

The light’s journey across space is the short part. The energy it carries was released by fusion in the Sun’s core, and reaching the surface takes far longer, because radiation is absorbed and re-emitted countless times on the way out. One estimate puts that at around 170,000 years ³.

Feynman’s examples were experiments that copied the outward form of science without its substance: the right rituals, but no honest test that could have proven them wrong.

The practical version is a habit of asking, before you share a conclusion: what would I expect to see if I were wrong, and did I look for it? The talk calls this “leaning over backwards” ¹. It is uncomfortable precisely because the person most motivated to miss the flaw is you.