Gregory Gundersen describes memorizing the first 100 digits of π as an exercise in engineering rather than a display of exceptional recall. He says he learned the sequence in roughly an hour after already building a person-action-object system, or PAO, and maintaining it with Anki. That qualification matters: the hour measured the final encoding, not the months of preparation required to create and retain the underlying mappings.
The method begins with elaborative encoding, which replaces abstract material with more memorable associations. In a PAO, every number from 00 through 99 is assigned a person, an action, and an object. A six-digit block combines the person from the first pair, the action from the second, and the object from the third. Gundersen’s example combines Albert Einstein, twirling, and a leotard to represent 502884. One hundred decimal digits therefore become seventeen six-digit images, with two digits left over.
Those images are then placed along a familiar route or inside a well-known building—the method of loci, commonly called a memory palace. Retrieval consists of mentally walking through the locations and decoding each scene. Support-2 describes the same spatial mechanism: information is associated with specific places, and recall follows the sequence of those places. That source supports the general method, but it does not independently confirm Gundersen’s personal timing, accuracy, or ability to reverse the sequence.
The main practical advantage is error locality. Gundersen reports that mistakes usually affect one glyph or misread one two-digit component by a decade, rather than causing him to lose his place in the entire recitation. Because each palace location has an independent cue, he can also pause and resume without reconstructing the sequence from the beginning. Walking through the palace in reverse provides a route for backward recall. These are his observations from one trained system, not controlled evidence that PAO prevents errors for every learner.
The hidden cost is system construction. Gundersen says the most time-consuming part was not reciting π but building a spreadsheet of memorable, distinctive people, actions, and objects. Similar entries can be confused, so the mappings must be sufficiently different from one another. He maintained those mappings during subway commutes because he expected the reusable number system to have value beyond this single feat. A beginner copying the technique should therefore budget for setup and review, not just the target list.
The apparent contrast with ordinary short-term memory also needs care. Gundersen invokes a capacity of five to nine objects, echoing Miller’s often-cited 7 ± 2 figure. Support-1 identifies that figure as a heuristic about chunks, not a universal memory limit, and notes that later research places working-memory capacity closer to four chunks under some conditions. Chunk size depends on familiarity and the material being stored. PAO works by making each encoded block meaningful enough to behave as a retrievable unit; it does not expand short-term memory in a simple, measurable way.
For readers considering a practical use, the method offers a clear design question: what makes recall recoverable after an interruption or a mistaken cue? Gundersen’s account suggests that spatial order, vivid associations, and independent checkpoints can answer that question better than a linear poem or unstructured repetition. It does not establish that memorizing π improves general intelligence, that the technique is equally efficient for every kind of information, or that an hour of practice is a realistic starting point. The evidence supports a narrower conclusion: a prepared mnemonic system can convert one narrow memorization problem into a disciplined encoding and retrieval workflow.