Remove one lesson. Keep everything else.
Step 1 of 4: Learned
One lesson removed, the rest kept.
Remove one outdated or unwanted lesson and keep everything else, with nothing to retrain.
Forgetting and data control. It runs On an authorised request.
- 40 / 40*Poisoned examples removed, with legitimate learning kept. See the result →
* Figures from Vareon’s internal benchmarks, not independently verified. Read the disclaimer
Outdated, wrong or unwanted: it goes.
A lesson can be removed because it is out of date, because it was wrong, or because it should never have been learned. You choose, and everything else stays.
Step 1 of 4: Out of date
Out of date: last year’s price list.
Nothing to retrain.
With Duplex, taking one lesson out takes one request. Taking one out of a trained model means training it again.
Step 1 of 3: Find it
First, find the lesson.
Learns and unlearns.
Duplex learns from each confirmed result straight away, and any single lesson can be taken back while the rest stays.
- It learns from every confirmed resultConfirm a result once, and the next request already benefits, with no retraining and no re-prompting.
- A correction changes the next answerYour team stops repeating the same correction: once accepted, it shapes every similar answer after it.
- A wrong change is taken backA lesson that made answers worse can be reversed, and the answers that worked before come back.
- One lesson removed, the rest keptRemove one outdated or unwanted lesson and keep everything else, with nothing to retrain.
- It learns what each action doesPlans grow more reliable over time, because Duplex learns what each action changed.
- A wrong observation is removedRemove an observation that turned out wrong, and every plan built on it is checked again.
See it 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.