Didactical · makers of LORE
Every training course ever written picks one explanation per idea and hopes. The learners it doesn’t reach get the same words again, slower. LORE was built the other way round: it carries several routes to the same fact, and when someone misses a question it teaches that point again by a different route — narrated, in a voice that sounds like a person, in the moment.
Then it records which explanation finally worked — for that person, and across the whole cohort.
What we do
The didactic method — one instructor, one lecture, one pace for everyone — does not survive contact with a workforce. LORE keeps the instruction and drops the single pace.
A manual, a guidance document, a standard operating procedure. PDF or text, up to 40 MB.
Lessons in order, with every question tied to the exact passage of the source it came from.
Your staff read the draft before it goes out — or choose to let a course publish on its own.
Each lesson narrated, in the browser on any phone or on Android (iPhone app coming soon), with every word highlighted as it’s spoken.
Every wrong answer is a named misconception, written from the passage that warns against it.
A new explanation on that exact point plays next — and which one worked is the thing worth knowing.
When it doesn’t stick
Re-explaining the same way, slower, is what a bad teacher does.
Each route is a different cognitive path to the same fact — not a restyled version of the same sentence. However an explanation is produced, it is checked against the approved script before it reaches a learner; anything the lesson does not say is flagged or refused.
Why the voice has to be this good
Hundreds of alternative explanations only exist as a product if narrating one costs nothing but machine time. Thirteen corrections rendered in twenty-three seconds — six times faster than real time — which is why a whole course’s adaptive layer is minutes of GPU rather than a studio calendar. And a voice a learner will hear that often has to be one they can stand for an hour.
| Hosted voice API | LORE | |
|---|---|---|
| Who the voice is | A stock voice from a shared library, or a clone you upload. Your learners may have heard it selling software last week. | Craig and Randi. Professional broadcast announcers, named, who agreed to teach this material in their voices. |
| What it learned from | Largely read-aloud corpora — audiobooks and studio reads. The model learns what reading sounds like. | Broadcast voice-tracking: working announcers performing to an audience. The model learns what talking sounds like. |
| Inflection and emphasis | Tuned per request, sentence by sentence. Stress lands where the model guesses, usually mid-sentence. | Trained on long segments, so phrasing carries across a paragraph — the emphasis falls on the word the sentence is about. |
| How the audio gets made | Chunk the script, call the API, stitch, listen for the seams. A human narrator means a booked session and an edit. | Approved text renders straight through on the first pass. No session to book, no second take, no punch-ins. |
| Where the narration is made | On the vendor’s servers. Your lesson scripts go with it. | On a GPU we control, against our own model weights. Approved scripts are narrated here, and finished audio stays on infrastructure we run. |
Listen
Randi reading section 1.7.3 of Applied Engineering Principles — shipboard ungrounded electrical systems — exactly as LORE built it. Two and a half minutes off the renderer in one pass; nothing re-recorded, punched in or cut. Listen for where she leans on a word.
“Your ship’s electrical system is ungrounded. So, no path to ground, right? … It sounds like solid logic. It is a deadly mistake.” Lesson script · drafted from the source, approved by a human
The same material, talked through
The rest of section 1.7 — what current does to a body, and the rules for changing a fuse — as a two-host episode. It closes by pointing at the lesson above: “Next episode: the myth about ungrounded systems that has gotten people killed.” Same manual, same approved script, second register. Three and a quarter minutes, unedited.
RandiAnd do not cut the wire.
CraigNever. The flash and the whipping ends make you the second victim.
Episode script · drafted from the source, approved by a human
Where the read comes from
Before it narrated a lesson, this voice read live radio — retail copy, forecasts, breaks. That is the performance the model was built from, and it is why the instruction above doesn’t sound read. Three reads, unedited.
“Doors open Saturday at nine a.m. sharp — and the first twenty-five buyers get a five-hundred-dollar gift card.” Broadcast read · one pass, no edit
“We’ll top out at eighty-four this afternoon, down to sixty-one overnight.” Broadcast read · one pass, no edit
“Brand new music, plus your chance to win tickets. Stick around.” Broadcast read · one pass, no edit
Sources: Applied Engineering Principles, chapter 1 · broadcast copy · professional voice talent · rendered on our own GPU
What it learns
Because the same idea is taught several ways, LORE can record not just that someone missed a point but which route finally reached them. Roll that up across a cohort and a raw pass rate separates into four very different situations that look identical from the outside.
01
The students who did well overall got this one wrong more often than the students who did badly. A miskeyed or ambiguous item can carry a perfectly unremarkable pass rate and be invisible to any amount of reading — and obvious to the statistic.
02
Failed first time, passed after the correction. The retry is carrying the module, which means the script has a defect with a specific address — this is the signal that turns an adaptive loop into a course-improvement loop.
03
Low pass rate, and the retry is no better. Nothing is broken here — the topic simply deserves more of the course than it is getting.
04
Everybody passes it. It distinguishes nobody, and it is taking up a slot that could be telling you something.
Two things we will not do with this data. We do not hide the sample size: item statistics are noisy below thirty responses and meaningless below ten, so every figure carries its n and anything under the floor is reported as provisional. A course change made on six answers is a guess with a decimal point on it. And we do not sort people into learning styles — the visual/auditory/kinaesthetic typology does not survive testing. What gets recorded is observed behaviour: which explanation this person responded to, on this concept.
Straight about the state of it
Send the manual you’d start with. A pilot is one document, a small group of learners, and a few weeks.
