This repository contains benchmark results and evaluation data for Telos, an experimental AI architecture built from scratch in Go.
Telos is a personal research project exploring alternatives to conventional neural network architectures. The code is not currently public.
A standard reasoning benchmark evaluating logical inference and deduction capabilities.
- proofwriter-results.csv — Raw evaluation results
Telos (Greek: purpose) is designed to run on constrained hardware—CPU only, limited RAM, no GPU—as a self-imposed architectural constraint.
Early results include:
- 400/400 random mazes solved optimally after training on a single maze (9999x9999 cells, 3x3 view)
- 26 cognitive benchmarks passed, covering analogical reasoning, few-shot learning, and self-correction
More benchmarks and documentation will be added as testing continues.
Questions or collaboration inquiries welcome via LinkedIn or GitHub issues.