Research

Papers, frameworks & open tools

We publish working papers early and keep the code, data, and annotation kits open so anyone can rerun the experiments. Results are reported as they stand, caveats included.

Working paper Status: in revision
Models: Qwen2-7B, Mistral-7B, Gemma-7B
Materials: open suite, corpora, results, annotation kit

Words Are Not Containers

The common picture treats a word as a box with a meaning inside. This paper tests a different idea: that meaning is relational and lives in how a word is used. We look inside open transformer models to see whether their internal states track that difference.

  • Cross-model replication. The core signals, including model "fingerprints" and an idiom-vs-literal separation, replicate across three independent 7B model families.
  • Semiotic accuracy. Internal representations separate figurative from literal use well above chance in every model tested, and the effect is statistically strong in all three.
  • Human baseline. A blind human annotation study is under way so the model results can be compared against human judgment on the same items.
  • Open questions. Some follow-on experiments have only run on one model so far, and the layer-by-layer analysis is being reworked before we report it.

The full experiment suite, corpora, results, and annotation kit are public. Ask us for the repository link or a copy of the current draft.

Framework arXiv: 2411.15832
Zenodo: 10.5281/zenodo.19136396 (v2, March 2026)
Implementation: Headgate (in progress)

Open General Intelligence (OGI)

OGI is a modular architecture for general intelligence systems. Rather than one monolithic model, it coordinates specialized components through a governed fabric, with an executive gate that decides what runs, when, and on whose authority.

A follow-on paper addresses the mathematical rigor of the original framework and gives it a formal treatment and implementation. Headgate, an open reference implementation of the executive gater and fabric, is being prepared for public release.

How we work

Ground rules

  1. Show the code

    If a result can't be rerun from what we publish, we don't treat it as a result.

  2. Report it as it stands

    Negative results, bugs, and unreplicated findings get written down, not buried.

  3. Test across models

    One model is an anecdote. We replicate across model families before we generalize.

  4. Check against people

    Where it matters, model behavior gets compared to blind human judgment.