Company

We build AI for people who work with numbers, proofs and code.

ALSoftworks is an independent software company. Rosenblatt is our family of models for STEM and programming.

What we believe

Show the work

A model that explains its steps is easier to check, correct and trust. We train for that, not just for the final answer.

Measure honestly

We publish results when we have them, describe how we got them, and say plainly what isn't finished.

Match access to risk

Most of our models are broadly available. Our research tier, Mark, is limited to reviewed partners with extra safeguards.

Why “Rosenblatt”

The figure is a real perceptron, the learning rule Rosenblatt described in 1958, training in your browser. Click to add points; shift-click for the other class. Add points no line can split and you'll see the limit Minsky and Papert described.

epoch 1 · 0 mistakes
Fig. 1 A perceptron, learning live. Click the paper to add points (shift-click for the other kind), then watch it fix its mistakes.
  1. 1958

    Frank Rosenblatt describes the perceptron, a machine that learns from examples.

  2. 1960

    The Mark I Perceptron learns to recognise simple shapes through a 20×20 grid of photocells.

  3. 1969

    Minsky and Papert show the limits of a single layer, such as data no straight line can separate.

  4. Today

    Many layers, trained deep, turn the same idea into today's language models.

Work with us

Questions, partnerships or research access: start with a request and we'll get back to you.