Gorde:
| Egile nagusia: | |
|---|---|
| Formatua: | Recurso digital |
| Hizkuntza: | |
| Argitaratua: |
Zenodo
2025
|
| Gaiak: | |
| Sarrera elektronikoa: | https://doi.org/10.5281/zenodo.17873036 |
| Etiketak: |
Etiketa erantsi
Etiketarik gabe, Izan zaitez lehena erregistro honi etiketa jartzen!
|
Aurkibidea:
- <p><em>Eve</em> is a deterministic Hard Neuro-Symbolic (Hard NeSy) cognition engine designed as a certifiable alternative to black-box large language models. Instead of sampling from a neural prior, Eve maintains an explicit symbolic <strong>Truth Graph</strong> with epistemic modes and provenance, performs bounded hypothesis expansion with deterministic pruning, uses a truth-orthogonal neural scorer operating only on masked graph structure, and generates language through a deterministic paraphrase lattice (DR2) constrained by a segment-level projector.</p> <p>This preprint describes the Eve 0.2.0 architecture, its epistemic model, safety contract, symbolic operator family, DR2 realization system, and early empirical results demonstrating millisecond-scale reasoning and near-linear ingest scaling on synthetic workloads. Eve is intended as a reference design for <strong>certifiable, deterministic, audit-ready AI systems</strong> compatible with emerging regulatory frameworks such as the EU AI Act and NIST AI RMF.</p> <p>Author: <strong>Zachary Ford</strong>, Neocere Interactive LLC.</p>