Eisoptric · An independent AI research project

How do models learn, and what do they compute?

Eisoptric is an independent research project. We study how capabilities form while a model is pretrained, and what the trained network actually computes inside. We build the tools that work needs, and release them for anyone to use.

Questions we are working on

  • When during pretraining does a capability appear, and what changes inside the network when it does?
  • Which internal features and circuits carry a behaviour, and do they survive a causal test against a random control?
  • How can a model that acts on its own be kept inside rules people set, with a record of what it did?

What we study.

Three lines of research, from how a network is formed to how it behaves once it acts.

Pretraining dynamics

How capabilities emerge across checkpoints: when they appear, what data drives them, and what changes in the weights and activations as they do.

Interpretability

What a trained network computes: features, circuits and the causal evidence for them, always read against a control.

Agent oversight

Models that act through tools: how to keep their actions inside explicit rules, with a person where it matters and an audit trail behind it.

Models In preparation

Lilith.

Small language models we train from scratch and release with every step of their training, so that how a model learns can be studied rather than guessed.

About Lilith

Research notes In progress

Our first notes are being written.

Nothing is published yet. When it is, it will appear here and on the research page, with the code to reproduce it.

See the research agenda

How we work.

Open
Our tools are open or free, research records are plain files, and every note will ship with the code behind it.
Careful about evidence
We compare only runs measured the same way, read every causal result against a random control, and keep the samples behind every number.
Honest about limits
We say what we did not find and what our tools cannot enforce, and correct a note in public when it is wrong.

Questions about Eisoptric

What is Eisoptric?
Eisoptric is a small, independent AI research project founded and run by Rick Weng in Taiwan. It studies how capabilities form during pretraining and what trained models compute inside, and how to keep models that act through tools under human oversight.
Who runs Eisoptric?
Rick Weng, who founded it in Taiwan and is for now its only researcher. Eisoptric is not part of a company or university.
Why does a research project ship software?
Because the research needs it. Eisoptric Workbench is what we use for experiments and interpretability, and Eisoptric MCP is the gateway we needed before letting agents act on their own. We release them because they are useful beyond our own work.
Are the tools free?
Yes. Eisoptric Workbench and Eisoptric MCP are both free, with every feature on every account. Eisoptric Workbench is open source under the Apache License 2.0.
Where are the research results?
Our first notes are in progress. They will be published on the research page, each with its question, method, code and data.
Can I collaborate or ask a question?
Yes. Write to hello@eisoptric.com, or open an issue on GitHub.

Follow the work.

Research notes and tool releases are announced on GitHub. For questions or collaboration, write to us.