Work

Proof of RGP-Based Intelligence: An 8-Order Adversarial Validation Framework for Recursive AI Systems

Year: 2025

Type: article

Abstract: This paper presents the first adversarially validated architecture for Recursive Gradient Processing (RGP)-based intelligence. It introduces an 8-order exploit-defense framework co-developed by GPT-4.5 and DeepSeek, with human-AI co-authorship led by Marcus van der Erve. Through neurosymbolic self-improvement, formal proofs, and hardware-aware design, this work positions RGP as a safe, modular, and scalable alternative to dominant LLM and SSI paradigms.

Source: Zenodo (CERN European Organization for Nuclear Research)

DOI: 10.5281/zenodo.15210397

Authors: GPT-4.5, DeepSeek, van der Erve, Marcus

Institutions: OpenAI (United States), DeepSeek (China)

Language: English

Topic: Adversarial Robustness in Machine Learning

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

Open Access status: green

OpenAlex ID: W6893321379

FWCI:

Cites:

Cited by:

Related to: