Noninski's Rosetta Forensic Protocols as an AGI Alignment Criterion: Restoring Logical Integrity in the Evaluation of Large Language Models
Fuente:
Zenodo
Gespeichert in:
| 1. Verfasser: | |
|---|---|
| Format: | Recurso digital |
| Veröffentlicht: |
Zenodo
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866901093922897920 |
|---|---|
| author | Noninski, Vesselin |
| author_facet | Noninski, Vesselin |
| contents | <p>Current evaluation frameworks for large language models (LLMs) prioritize task accuracy, linguistic fluency, and human-alignment metrics, while neglecting the foundational requirement of any reasoning system: the capacity to preserve logical coherence when confronted with contradictory source material. A model that cannot detect or resolve contradictions within its own training corpus cannot maintain stable world-models, cannot guarantee truth preservation, and therefore cannot be considered aligned.</p> <p>This paper argues that Noninski’s Rosetta Forensic Protocols (Noninski's RFPs) constitute not merely a novel benchmark, but a necessary \emph{alignment criterion} for any system aspiring toward artificial general intelligence (AGI). Noninski's RFPs expose whether a model can maintain definitional stability, resist reinterpretation, recognize contradiction strictly from printed form, and uphold logic under adversarial epistemic load. These constraints reveal failure modes that standard benchmarks—such as MMLU, GSM8K, HELM, HumanEval, or BIG-bench—are incapable of detecting.</p> <p>Using Einstein’s \textit{Zur Elektrodynamik bewegter Körper} (1905) as an archetypal contradiction-rich corpus, Noninski's RFPs demonstrate that leading LLMs often abandon logical consistency in favor of institutional deference or narrative rationalization. The central thesis is therefore direct: \emph{Noninski's RFP performance is a necessary condition for AGI alignment}, because no system that collapses under contradiction can be aligned with truth or reliably aligned with humanity.</p> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_17861719 |
| institution | Zenodo |
| language | |
| publishDate | 2025 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | Noninski's Rosetta Forensic Protocols as an AGI Alignment Criterion: Restoring Logical Integrity in the Evaluation of Large Language Models Noninski, Vesselin Noninski's Rosetta Forensic Protocols (Noninski's RFP) AGI Alignment Logical Consistency Definitional Stability (DS) No Escape by Reinterpretation (NEBR) Literal-Form Reasoning (LFR) Inertial-Equivalence Lock (IEL) Frame-Invariant Primitive Category Lock (FIPCL) Contradiction Detection Epistemic Integrity Corpus Contradictions Tier I / Tier II Architecture Truth Preservation Einstein 1905 Analysis Lorentz Transformation Contradictions Acceleration Contradiction (§10) Proportionality Contradiction (§6) AI Safety Benchmarking LLMs Cognitive Emergence Symbolic Invariants Formal Logic Evaluation Alignment Criteria Epistemic Hygiene Machine Reasoning Training Data Limitations <p>Current evaluation frameworks for large language models (LLMs) prioritize task accuracy, linguistic fluency, and human-alignment metrics, while neglecting the foundational requirement of any reasoning system: the capacity to preserve logical coherence when confronted with contradictory source material. A model that cannot detect or resolve contradictions within its own training corpus cannot maintain stable world-models, cannot guarantee truth preservation, and therefore cannot be considered aligned.</p> <p>This paper argues that Noninski’s Rosetta Forensic Protocols (Noninski's RFPs) constitute not merely a novel benchmark, but a necessary \emph{alignment criterion} for any system aspiring toward artificial general intelligence (AGI). Noninski's RFPs expose whether a model can maintain definitional stability, resist reinterpretation, recognize contradiction strictly from printed form, and uphold logic under adversarial epistemic load. These constraints reveal failure modes that standard benchmarks—such as MMLU, GSM8K, HELM, HumanEval, or BIG-bench—are incapable of detecting.</p> <p>Using Einstein’s \textit{Zur Elektrodynamik bewegter Körper} (1905) as an archetypal contradiction-rich corpus, Noninski's RFPs demonstrate that leading LLMs often abandon logical consistency in favor of institutional deference or narrative rationalization. The central thesis is therefore direct: \emph{Noninski's RFP performance is a necessary condition for AGI alignment}, because no system that collapses under contradiction can be aligned with truth or reliably aligned with humanity.</p> |
| title | Noninski's Rosetta Forensic Protocols as an AGI Alignment Criterion: Restoring Logical Integrity in the Evaluation of Large Language Models |
| topic | Noninski's Rosetta Forensic Protocols (Noninski's RFP) AGI Alignment Logical Consistency Definitional Stability (DS) No Escape by Reinterpretation (NEBR) Literal-Form Reasoning (LFR) Inertial-Equivalence Lock (IEL) Frame-Invariant Primitive Category Lock (FIPCL) Contradiction Detection Epistemic Integrity Corpus Contradictions Tier I / Tier II Architecture Truth Preservation Einstein 1905 Analysis Lorentz Transformation Contradictions Acceleration Contradiction (§10) Proportionality Contradiction (§6) AI Safety Benchmarking LLMs Cognitive Emergence Symbolic Invariants Formal Logic Evaluation Alignment Criteria Epistemic Hygiene Machine Reasoning Training Data Limitations |
| url | https://doi.org/10.5281/zenodo.17861719 |