Skip to content
Advancing Bits to Master Atoms

Continual learning.
Your AI learns from every confirmed result.

Today’s models stop learning on the day their training ends. MS Duplex keeps the AI your scientists and engineers use learning as the work goes on.

Learn and improve · one of the fourteen capabilities of MS Duplex

An upright loop of satin-silver ribbon on a graphite plinth, a growing row of small cobalt-blue beads set along its inner edge.An upright loop of satin-silver ribbon on a graphite plinth, a growing row of small cobalt-blue beads set along its inner edge.

Your team gets

  • Your AI keeps up with your research, every day it runs
  • No training runs, datasets to prepare or accelerators to book
  • Earlier lessons stay sharp as new ones arrive

What continual learning means for your team.

Continual learning means your AI keeps learning after it is deployed, from the work your team does. Every confirmed outcome, every expert correction and every approved conclusion becomes experience the next request can use.

A model on its own cannot do this. Its knowledge is fixed when its training ends, and teaching it something new means preparing a dataset and running another training cycle. MS Duplex learns beside the model instead, with the model’s weights left as they are, so a lesson your team confirms this morning applies this afternoon.

Learning more does not mean forgetting. A new lesson is added to what MS Duplex already holds and leaves the rest as it was, so the AI that learned your newest method still knows the conditions your team confirmed months ago.

Why research needs AI that keeps learning.

Science moves on every day. A model frozen at the end of its training falls further behind your work with every week it runs.

In a research team, the most valuable knowledge is the newest: the measurement that came back yesterday, the correction an expert made this morning, the operating limit someone confirmed last week. None of it is in any model’s training data, because it did not exist when the model was trained.

Without continual learning, that knowledge lives in people’s heads and in notes the AI never reads properly. Experts repeat the same correction, agents propose the same failed step again, and runs that cost real compute and lab time are spent rediscovering what the team already knew.

  • Knowledge that is days old

    The results that matter most to your next decision are the ones your own team produced most recently.

  • Expert time is scarce

    A correction made once should not have to be made again by the same expert in the next session.

  • Runs are expensive

    Every repeated failure costs compute, materials or instrument time that a learning AI would have saved.

  • Facts change

    A newer measurement replaces an older one. Your AI has to move on with it, not keep quoting the old value.

Four small modules on a round graphite base, a graphite block, a silver gauge, a silver tray and a cobalt-blue tablet, joined in a ring by one thin cobalt-blue track.Four small modules on a round graphite base, a graphite block, a silver gauge, a silver tray and a cobalt-blue tablet, joined in a ring by one thin cobalt-blue track.

How continual learning works in your team’s day.

Your team keeps working the way it does today. MS Duplex learns from the outcomes you choose to capture and brings them back where they fit.

  1. 1

    Your agent works

    It plans, runs a simulation, analyzes data or proposes the next step, with the model and tools your team already uses.

  2. 2

    The result is confirmed

    A test passes, a measurement comes back or an expert approves the conclusion, or corrects it and says why.

  3. 3

    The lesson is kept

    MS Duplex records what was learned with its evidence and the conditions it holds under, ready from the next request.

  4. 4

    It applies where it fits

    The next matching case, in any session, agent or model, starts from that lesson. Where conditions differ, MS Duplex says so.

An example, in any field where an agent helps decide.

A team’s agent recommends settings for a recurring analysis. Over one week, it learns three things nobody had to write down twice.

  1. 1

    A correction

    On Monday an expert rejects a recommendation because it ignores a known limit of the method, and records the reason. From the next request, the agent checks that limit first.

  2. 2

    A confirmation

    On Tuesday a result the agent predicted is confirmed by measurement. The approach behind it becomes the agent’s first choice for cases with the same conditions.

  3. 3

    A changed fact

    On Thursday a new calibration replaces an old reference value. Every later answer uses the new value, and the history shows when and why it changed.

  4. 4

    The next session

    On Friday a colleague opens a new session on a different model. All three lessons are already there, each with its source.

Continual learning makes the other capabilities stronger.

What MS Duplex learns continually is what the rest of the layer builds on.

  • Self-improvement

    Lessons learned continually become better behavior: what worked is tried first, what failed stops being proposed.

  • Reusable methods

    A confirmed procedure becomes a method your agents reuse and adapt, the moment it is confirmed.

  • Research memory across sessions

    Each new session starts from everything learned so far, not from what the model knew at the end of its training.

  • Provenance, review and unlearning

    Every lesson keeps its source, so any one can be reviewed, corrected or removed without touching the rest.

What your team gets.

AI that keeps up with your research, with every lesson visible and under your control.

answers about a changed fact gave the outdated value, in MatterSpace’s controlled tests
0 of 8,640
stale answers after facts changed: MS Duplex beside standard retrieval, in the same tests
0% vs 57%
  • Learning from the next request

    A confirmed result or correction applies from the next request, with no training cycle to wait for.

  • No training runs

    No datasets to prepare, no accelerators to book and no new model version to validate after every lesson.

  • Nothing forgotten

    A new lesson leaves what was learned before intact, so older results stay as useful as newer ones.

  • Current facts

    When a value changes, the newest confirmed one applies and the history of the change is kept.

  • Your choice of what is learned

    Your team decides which outcomes and corrections are captured, and can review every lesson.

  • Any model

    The learning sits beside the model, so it works with a general model, a fine-tuned one or the next release.

Common questions.

Does continual learning change my model?

No. The model’s weights stay as they are. MS Duplex learns beside the model, which is why a lesson is ready from the next request and why the experience stays when you change models.

What does MS Duplex learn from?

From the outcomes and corrections your team chooses to capture: a passed test, a measurement, an approved conclusion, an expert’s correction with its reason. Each lesson keeps its source and the conditions it holds under.

Will new lessons make it forget older ones?

A new lesson is added beside what MS Duplex already holds and leaves the rest as it was. Where a newer confirmed result replaces an older one, the newer one applies and the history is kept.

How is this different from fine-tuning?

Fine-tuning changes a model’s weights through a training run on a prepared dataset. Continual learning in MS Duplex takes one confirmed result at a time, applies it from the next request and keeps every lesson editable. In MatterSpace’s controlled tests, MS Duplex came out ahead of a model fine-tuned for the task on recurring work.

Make the next run benefit from the last.

Bring us the research your AI should get better at. We will plan a pilot on your own workflows, with your experts as the judges.