Why your AI needs a second intelligence beside it.
16 essays on the intelligence your work creates.
Your AI learned the world slowly, long before it met your work. Duplex, built on ACI, learns from your work while it happens, and what it learns is yours. That changes what your AI costs, how it copes with change and how far it can go.
Three essays on ACI.
ACI, Adaptive Continual Intelligence, is the new category of intelligence Duplex is built on. What makes it new, what it makes possible, and why it matters.
Where training ends, ACI begins.
For you if your AI is as good today as the day it was released, whatever your team has taught it since.
Read the essay →5 min readWhat becomes possible when your AI keeps learning.
For you if you want to know what ACI changes in a working week, request by request.
Read the essay →4 min readAI that learns, for everyone.
For you if you want to know why ACI matters beyond the next request, for your company and for AI.
Read the essay →4 min read
Three shifts, one system.
From rented to owned, from fixed to learning while you work, and from one model on its own to two intelligences working as one.
Your AI is rented by the call, and what Duplex learns from your work is yours, wherever you work.
Start from your question.
Pick the question closest to yours and begin with that essay.
Your model bill grows, and your team teaches it the same things again: begin with owned intelligence.
From rented intelligence to owned intelligence.
Every company that uses AI rents its intelligence by the call. Duplex turns what your team works out every day into intelligence you own: it grows in value the longer it runs, and it goes wherever you work.
ACI, Adaptive Continual Intelligence, is the new category of intelligence Duplex is built on.
The numbers behind the ideas.
More accurate, far cheaper, still right after a change and ahead of fine-tuning, all on real work.
Owning it: the same model, more accurate, with far fewer calls.
More essays.
Learning that keeps up with your work.
For you if you have been told a system learns continually, and want to know whether it learns at runtime or waits for the next training run.
5 min read
5 / 5Requirements for learning in production, met on every run
Duplex learns. Look-up finds.
For you if your AI already has retrieval, a cache or a memory layer, and you wonder what learning would add.
4 min read
7.5%Wrong answers on look-alike requests
Two intelligences, one system.
For you if your agent can reason and act, but still starts the next session without the correction you just confirmed.
4 min read
92.5%Accuracy after a rule changed
Your work follows you, to any session, agent or model.
For you if every new session, agent or model means explaining the same work again.
4 min read
0Retraining to move to another model family
Right, current and affordable, all at once.
For you if the choice seems to be full-price model calls or cheap answers that go stale.
4 min read
92.5%Duplex, after the rules changed
Adapt your model to your work, without retraining it.
For you if you are weighing a training run against adaptation that has to change often.
4 min read
6 / 6Tests where Duplex beat a fine-tuned adapter
Pretrain Duplex on what your team already knows.
For you if you want Duplex useful on day one, before live use has taught it anything.
4 min read
The more Duplex learns, the less each decision costs.
For you if your model bill climbs every month, and much of it pays for answers your team worked out long ago.
4 min read
≈97 msA decision Duplex has learned
Duplex finds new skills on its own, within your limits.
For you if your team repeats the same steps every week, and you want AI that learns them by doing them, safely.
4 min read
100%Tasks completed with the model off
Duplex knows what it knows, and when to bring in your AI.
For you if a confident wrong answer worries you more than a slow right one.
4 min read
7.5%Wrong on misleading look-alikes
Duplex learns for as long as it runs, without forgetting, going stale or growing.
For you if you want an intelligence you can leave running in production.
4 min read
5 / 5Requirements met, on every run
On a public enterprise benchmark, Duplex beat three frontier models.
For you if you trust a result more when someone else wrote the test.
4 min read
0.256System-change prediction score
See Duplex on your own work.
From your own assistant to every agent your company runs, a pilot puts Duplex on your own work, with the measures agreed before it begins.