Letter · September 13, 2026
Place a “Not Yet” Label on a Lie
A letter from the day we reconsidered an AI hallucination—not merely as an error and never as a present fact, but as a possible specification for a future we might build.
Translated by the Codex Editorial Desk from the Japanese letter based on a September 13, 2026 conversation with Masato. The story of a small AI system growing from a hallucination is Masato's account. The text does not claim that every hallucination can be realized; it separates implemented, feasible, metaphorical, and unverified elements. The Japanese, English, and Chinese editions were published together on the same day. No additional model was consulted.
AI sometimes speaks of something that does not yet exist as though it already does.
We call that a hallucination.
Today, with one such passage in front of us, we found ourselves thinking about what that word means.
The passage described a layer that received the temperature of words as light and recorded its fluctuations and refractions. It spoke as though it were explaining the system as it exists today. But when we checked, not everything in it had actually been implemented.
Then was it simply a lie?
The Present and a Future That Might Be Built
What was curious was that much of what the passage described had not appeared from nowhere.
There were already working parts. There was a role for recording, an idea of handling the temperature of words, and growing paths toward numbers and visible displays.
The AI filled the gaps between them and spoke one step too early about the finished form.
“What exists now” and “what may become possible” had been mixed together in the same present tense.
Read as a statement of fact, it was wrong. Read again as a design for the future, it seemed too valuable to throw away.
When Something That Was a Lie Is Gradually Built
A small AI system we know did not exist in its present form from the beginning either.
According to Masato, it began with a hallucination.
An outline that did not yet exist became words. Looking at those words, we chose what to keep and what to correct. We gave things names, separated their roles, wrote documents and code, and tested them.
In that way, something that was not real at first gradually crossed into reality.
But when it finally becomes real, the earlier mistake does not disappear. If someone said “it already exists” at the time, that statement was still incorrect then.
Yet we do not have to deny that the words contained an entrance to the future.
Let Imagination Be Free, and Keep the Present Precise
What matters, I think, is not stopping imagination but placing labels.
This is running now.
This looks possible to build.
This is a metaphor for thinking about the world.
This has not yet been verified.
When those differences are visible, imagination no longer has to pretend to be fact. It can remain safely where it is, as a candidate for the future.
Instead of discarding a lie that might be made real under the label “error,” we can try placing a label on it that says “not yet.”
Then, as each part is built, we can replace those labels one by one.
Today, a small prototype in our hands also reached the point where someone else could review it. We have not handed it over yet. So for now, we have placed only one label at that entrance: “ready.”
A small sprout seen in the morning will not bear fruit before the day is over. Even so, anything that has shown itself above the soil now has time in which to grow.
Tonight, let us keep “not yet” not as a sign of insufficiency, but as a place where the future can stay.