Trie Memory — Epoch Reading

▲

The lines he pulled

“The question of efficiency of space utilization is one with which we shall be primarily concerned.”

p.491 · starred as the pillar — the thesis, found on page two

“The main disadvantage is relative inefficiency in using storage space, but this inefficiency is not great when the store is large.”

p.490 · margin: “context limits in chat” — the analogy that opened and closed the read

“The branching, tree-like structure is a result of an attempt to simplify the diagram; in principle, directly connected registers need not be physically near each other.”

p.494 · margin: “likely looks more like a map — Obsidian”

“In a computer designed around the trie memory principle, one would have only to name the procedure and the data to which it should be applied.”

p.498 · margin: “the birth of the vibe code — EPOCH OS”

“In the trie memory scheme, the information is filed only once, not in multiple.”

p.498 · single-filing against human recall — “faint memories”

●

The read, page by page

p.490

AThe advantages list underlined — access, updating, diverse lengths, redundancy — read as the ability to hold and cross-reference.
B“The store is large” → context limits in chat. — the modern wall named on page one of a 1960 paper.
CThe register grid read as a battleship board. — coordinate addressing, compressed to a game.

p.491

★The space-utilization question starred as the pillar. — thesis located, two pages in.
BIs crossing through the blank space a barrier or drag feature of this design? — the right question; the paper's answer is space, not speed. Reads never touch the blanks.

p.493

BThe deletion register: a scrubber worm, or a read-already caching thing? — half right, and the right half: functional deletion is cache-invalidation, 1960 edition.
DDe la Briandais read correctly as run-through-each-to-match. — the space-for-time trade Fredkin refuses.

p.494

BTree diagram called what it is: a map, an Obsidian graph. — Fredkin says himself the tree is only a drawing convenience.
C“The whole game here / the jist of the problem” bracketed at the N-dimensional turn. — the paper's hinge, found in real time.
DNeighbor allocation mapped onto the live model landscape. — his running analogy for how the big labs differ on memory.

p.495

BLevels filling and spilling — “like how Bohr models work?” — shells to capacity, sparse at the rim. It holds.
FTrapping condition read as “until context was full and no new space to pack.” — the morning analogy carried through the Monte Carlo math.

p.496

ABigger trie, bigger capacity, more connections within. — the paper's own closing extrapolation, restated cold.
CFig 7A vs 7B split: scale assessment versus quality assessment. — exactly the split.

p.497

CThe “ing” compaction scheme called “shorthand for compiler paths.” — replace the n-gram with a pointer to a model sequence. That is a token, spotted 60 years before the word.

p.498

EName the procedure and the data, and the program assembles: “the birth of the vibe code — EPOCH OS.”

p.499

AApperceptive memory bracketed: “is this what agent frameworks are doing?” — pre-fetch from slow storage, retrieve the most-nearly-relevant. Yes, structurally.
★Done @ 1:33 P.M., in ink, at the references — where citation 3 is Licklider. — the next read was already on the desk.

Trie MemoryEdward Fredkin · Bolt Beranek and Newman · Communications of the ACM, 1960 · pp. 490–499

The constellation the margins drew

The lines he pulled

The read, page by page