Listing by conversation
Can a home be listed through a chat, without a single form?
Loading…
Experiments
Prototypes and open questions.
Experiments
Things we are trying, shown honestly.
Experiments is where we publish prototypes and the questions behind them. Some become products or research; many do not. Each is labelled with what it is, what we learned and what is still unknown.
Current experiments
Saying how sure it is
Can a system show its confidence in words a person reads at a glance?
Knowing the neighbourhood
What should an agent know about a place before it suggests what to do there?
How we run them
01Start with a question
Each experiment begins as a plain question, like the three above, not as a feature waiting to ship.
02Build the smallest thing
A prototype just large enough to show whether the question has a useful answer.
03Write up what happened
What it is, what we learned and what is still unknown — including when the answer turned out to be no.
Where it shows up
- 01From a message to a listing someone can checkExperiment · In progress. Listing by conversation, written up as it runs.
- 02Trust ScoreSaying how sure a system is sits close to showing the reasons behind a score.
- 03Spatial IntelligenceKnowing the neighbourhood is a question about what an agent should be told about a place.
- 04ResearchEvery note and experiment, labelled with how far the work has got.
What a person checks
01
What the prototype showed
A demo that worked once is not a result. Someone checks what it did across the cases that matter.
02
The label
Each experiment says whether it is in progress, conceptual or published, and the label is kept honest.
03
Whether to stop
Many experiments do not become products or research. Deciding to stop is part of the work.
Read the write-ups.
Research notes and experiments, published as the work produces them.
