Research
Advancing the science of one more sentence.
Our researchers publish a paper with every release. Publication follows training. Peer review remains pending.
Technical report
brAIAIn: A Language Model Trained Exclusively on What It Was Paid to Learn
BRAIAIN Laboratories, November 2026. Method, every version's measurements, and the complete training data, in five pages. Peer reviewed by nobody. Permanently archived anyway.
Read the PDFDOI: 10.5281/zenodo.23863739
Cite this report
@techreport{braiain2026,
title = {brAIAIn: A Language Model Trained Exclusively on What It Was Paid to Learn},
author = {{BRAIAIN Laboratories}},
institution = {BRAIAIN Laboratories},
year = {2026},
doi = {10.5281/zenodo.23863739},
url = {https://doi.org/10.5281/zenodo.23863739}
}Interpretability
Tracing the thoughts of a language model, completely.
Other labs are building microscopes to look inside their models. Ours does not need one. Every word brAIAIn chooses is drawn from the counts of what followed the previous two words, and all of the counts are public.
Loading the circuit.
Planning ahead: none found. It picks each word after the one before it, like a person who started a sentence without knowing how it ends. Read every count in latest.json.