We publish when there are results.
This list is short on purpose. Three programmes are running; several more are not far enough along to be worth anyone's attention yet, and we would rather have an empty page than a speculative one.
In research
Axiom
How advanced intelligence converges toward stable structure as complexity increases.
Foundational research
Restricted
MG-SGLoRA
Steering a trained model without retraining it — sparse to apply, and reversible.
Model control
Restricted
CRSTO
Behaviour-aware compression: shrink a model without losing the discipline that made it useful.
CompressionHow we handle unfinished work.
Several programmes here have no page because they have no results. When they do, the result goes up whether or not it went the way we hoped — a claim that fails its own test is still a finding, and burying it would make every other number on this site worth less.