Artificial Intelligence

Beyond the Data: Analysis, Feature Engineering and Browser Plugin Expansion for the Sharelm Dataset

Authors: Samer Attrah
Comments: 9 Pages. 3 tables, 5 figures

As part of the Eleuther AI open AI summer research this year, we worked on expanding the ShareLMdataset browser extension, by adding support to multiple models in addition to redesigning some ofthe visual parts of the extension, in the mean time conducted several analysis and feature engineeringon the ShareLM dataset to extract insight regarding the models, users, the conversations and therelations connecting them.
Category: Artificial Intelligence

Analogy as the Core of Intelligence

Authors: Akira Pyinya
Comments: 22 Pages.

This article argues that the core of intelligence is not optimization, but analogy. We define intelligence as "doing the same thing as the examples of the right thing to do in new situations." We transform Hofstadter's Copycat problem into a sequence prediction problem to derive a formal definition of analogy-based intelligence, from which value functions and temporal-difference error can be derived, showing that optimizers can be derived from analogy-based systems. We demonstrate how agency and free will arises from conflicts between different predictions based on different examples.
Category: Artificial Intelligence

Red Teaming Quantum-Resistant Cryptographic Standards: A Penetration Testing Framework Integrating AI and Quantum Security

Authors: Petar Radanliev
Comments: 33 Pages.

This study presents a structured approach to evaluating vulnerabilities within quantum cryptographic protocols, focusing on the BB84 quantum key distribution method and National Institute of Standards and Technology (NIST) approved quantum-resistant algorithms. By integrating AI-driven red teaming, automated penetration testing, and real-time anomaly detection, the research develops a framework for assessing and mitigating security risks in quantum networks. The findings demonstrate that AI can be effectively used to simulate adversarial attacks, probe weaknesses in cryptographic implementations, and refine security mechanisms through iterative feedback. The use of automated exploit simulations and protocol fuzzing provides a scalable means of identifying latent vulnerabilities, while adversarial machine learning techniques highlight novel attack surfaces within AI-enhanced cryptographic processes. This study offers a comprehensive methodology for strengthening quantum security and provides a foundation for integrating AI-driven cybersecurity practices into the evolving quantum landscape.
Category: Artificial Intelligence

Generative Models Enable True Understanding: The Link Between Interpretability and Generative Ability

Authors: Yuan-Hao Wei
Comments: 9 Pages.

Interpretability and generative capability in generative models are fundamentally two complementary aspects. A highly interpretable model typically learns the true underlying generative mechanisms behind data, such as physical laws, causal relationships, or explicit structures. As these mechanisms are inherently stable and universally applicable, such models can reliably generalize beyond training data, producing more reasonable and robust samples with fewer generation failures. In addition, a highly controllable and powerful generative model implicitly or explicitly captures genuine and effective underlying rules. The ultimate goal of training generative models should extend beyond obtaining high-quality samples to exploring and understanding the underlying generative mechanisms of phenomena. When a generative model demonstrates controllability and scalability with respect to a dataset, it indicates the model has genuinely learned the mechanisms that generate the data. This opens up a paradigm in scientific research, enabling the discovery of underlying principles through observational data reconstructed by generative models, particularly when these models exhibit controllability and scalability. Leveraging powerful nonlinear mapping, efficient iterative training, and structured interpretability, artificial intelligence holds the potential to uncover and understand rules and principles currently beyond human knowledge.
Category: Artificial Intelligence

Unified Framework for Efficient Cross-Lingual Transfer Learning Across Low-Resource Languages Using Knowledge-Augmented Multilingual Models

Authors: Ritika Budhiraja, Bhaumik Tyagi, Sagar Kumar Jha
Comments: 9 Pages.

Cross-lingual transfer learning is incredibly promising for facilitating knowledge transfer between languages, particularly for low-resource languages that lack annotated data. However, many current methods are inefficient in terms of adaptation, have poor generalizability, and often fail to incorporate external real-world or linguistic knowledge. This paper introduces a Unified Framework for Efficient CrossLingual Transfer Learning Across Low-Resource Languages using Knowledge-Augmented Multilingual Models. The approach integrates structured and unstructured knowledge sources, such as multilingual knowledge graphs, lexical resources, and cross-lingual embeddings, into pre-trained multilingual language models (like XLM-R and mT5) through adapter-based fine-tuning and prompt-guided alignment. This creates a task-agnostic transfer pipeline that jointly optimizes for semantic alignment, knowledge consistency, and lowresource adaptability across multiple NLP tasks, including machine translation, named entity recognition, and question answering. Experimental results on 25 typologically diverse languages, including some with fewer than 10,000 training examples, demonstrate that the framework achieves state-of the-art performance, significantly surpassing current multilingual baselines in zero-shot and few-shot regimes. Furthermore, ablations reveal the critical contribution of knowledge integration to improving contextual disambiguation and representation fidelity for low-resource languages, providing a foundation for creating scalable, knowledge-driven multilingual systems that help close the digital linguistic divide.
Category: Artificial Intelligence

Spectral Analysis of State Space Models in Language Modeling: Training Dynamics and Stability Properties

Authors: Zayan Hasan, Aniketh Malipeddi, Aneesh Chatrathi
Comments: 5 Pages. (Note by viXra Admin: Please submit article written with AI assistance to ai.viXra.org)

State Space Models (SSMs) have emerged as a new linear computational complexity transformer rival to sequence modeling on long sequences with competitive performance. The dynamics of training and stability properties of SSMs remain poorly understood from a spectral perspective. This work presents the first wide reaching spectral analysis of SSM based language models. Providing a systematic framework to examine how eigenvalue distributions and spectral radii evolve during training, through experiments on a 737K parameter SSM model having 3 layers, state space dimension 128, and model space dimension 8, it was discovered that although the minority of the state matrices learned lead to theoretical spectral stability with mean spectral radius 1.078. The model demonstrates excellent convergence however, reducing training loss from 3.127 to 0.305 using 100 epochs. The eigenvalue analysis demonstrates common clustering in the negative real axis with concentration centered about negative 0.8, exhibiting a bimodal spectral radius distribution exhibiting systematic behavior in SSM dynamics. The key result portrays that SSMs operate efficiently in scenarios such as these. The selective mechanism provides adaptive control that prevents mathematical instabilities from causing training divergence. This renders assumptions of classical neural network stability hard to maintain and makes spectral analysis an essential for understanding similar model behavior. This work provides practical insight toward constructing more principled, stability aware designs for such models and frameworks.
Category: Artificial Intelligence

AI Above: Securing Aviation with Intelligent Systems

Authors: Mezbah Uddin Rafi
Comments: 17 Pages. (Note by viXra Admin: Please cite listed scientific reference and submit article written with AI assistance to ai.viXra.org)

The sky has always been a symbol of freedom, progress, and limitless possibility—but with each advancement in aviation, new risks emerge that challenge our ability to keep flight both safe and secure. Today, as global air travel surges and flight systems grow increasingly complex, the aviation industry turns to a new co-pilot: Artificial Intelligence (AI). No longer a speculative technology, AI is actively reshaping how we safeguard passengers, crews, aircraft, and infrastructure from both traditional dangers and modern threats. This paper embarks on an in-depth exploration of AI’s transformative role in aviation security and accident prevention. From intelligent surveillance and predictive diagnostics to autonomous flight corrections and cyber threat mitigation, AI systems are revolutionizing every stage of aviation operations. Machine learning models, trained on vast datasets of flight telemetry and maintenance records, now predict component failures before they occur. Neural networks embedded in cockpit systems assist pilots with real-time decision-making during critical scenarios, while AI-powered air traffic control systems optimize flight paths, reduce congestion, and enhance mid-air conflict resolution. Furthermore, biometric authentication and behavioral analytics are reinforcing aviation security at a human level—preventing unauthorized access and identifying suspicious activities with unprecedented accuracy. But alongside the benefits come profound ethical and regulatory questions. Who holds accountability when AI intervenes—or fails—in the flight deck? How do we balance autonomy and human oversight? This paper also unpacks the societal and legal implications of AI integration in aviation, including concerns over data privacy, algorithmic transparency, and the digital divide between nations with differing technological capacities. Through recent case studies, ongoing trials by aerospace leaders, and insights from interdisciplinary research, this study builds a comprehensive picture of AI as a guardian of the skies. It illustrates how intelligent systems are evolving beyond supportive tools into autonomous protectors—capable of adapting, learning, and responding in ways that enhance resilience, reduce error, and fortify aviation against tomorrow’s unknowns. In an age where every flight carries the weight of both human dreams and global risk, Artificial Intelligence offers a path forward: one that is safer, smarter, and fundamentally more prepared to meet the boundless challenges of modern aviation.
Category: Artificial Intelligence

GSPNN: Graph Shortest Path Neural Network

Authors: Atharv Navale
Comments: 5 Pages. (Note by viXra Admin: Please submit article written with AI assistance to ai.viXra.org)

A neural architecture termed GSPNN (Graph Shortest Path Neural Network) is introducedin which both inference and learning are carried out without backpropagation. Classificationis posed as a shortest-path problem over a layered directed acyclic graph (DAG). Each layerdefines a local softmax distribution over outgoing edges; per-edge cost is the (temperaturescaled) negative log-probability. Inference reduces to a Viterbi (min-sum) dynamic programover the graph. Learning proceeds by local, forward-only updates: for each training example,the shortest path to the true class is compared with the best competing class. If a margin isviolated, normalized per-node updates are applied that increase the probability of chosen (true)edges and decrease it for competing (wrong) edges. The updates require only forward signals(local probabilities and keys) and avoid gradients and backpropagation. On MNIST withPCA features, a compact configuration attains 94—96% test accuracy with sub-second epochtimes on a single Colab GPU due to fully vectorized updates and optional precomputation of per-layer keys. Algorithmic details, computational complexity, ablations (temperature and margin schedules, top-k negatives, EMA), limitations, and connections to Viterbi decoding, structured prediction, and local learning rules are discussed.
Category: Artificial Intelligence

A Multi-Component AI Framework for Computational Psychology: From Robust Predictive Modeling to Deployed Generative Dialogue

Authors: Anant Pareek
Comments: 8 Pages. (Note by viXra Admin: Please submit article written with AI assistance to ai.viXra.org)

The confluence of Artificial Intelligence and Computational Psychology presents an opportunity to model, understand, and interact with complex human psychological states through computational means. This paper presents a comprehensive, multi-faceted framework designed to bridge the gap between isolated predictive modeling and an interactive system for psychological analysis. The methodology encompasses a rigorous, end-to-end development lifecycle. First, foundational performance benchmarks were established on four diverse psychological datasets using classical machine learning techniques. Second, state-of-the-art transformer models were fine-tuned, aprocess that necessitated the development of effective solutions to overcome critical engineering challenges, including the resolution of numerical instability in regression tasks and the creation ofa systematic workflow for conducting large-scale training under severe resource constraints. Third, a generative large language model (LLM) was fine-tuned using parameter-efficient techniquesto function as an interactive "Personality Brain." Finally, the entire suite of predictive and generative models was architected and deployed as a robust, scalable microservices ecosystem. Keyfindings include the successful stabilization of transformer-based regression models for affective computing, showing meaningful predictive performance where standard approaches failed, and the development of a replicable methodology for democratizing large-scale AI research. The significance of this work lies in its holistic approach, demonstrating a complete research-to-deployment pipeline that integrates predictive analysis with generative dialogue, thereby providing a practical model for future research in computational psychology and human-AI interaction.
Category: Artificial Intelligence

The Geometry of Forgetting: Toward a Law of Information Decay in Self Modifying Systems

Authors: Jace Hall
Comments: 15 Pages. (Note by viXra Admin: Please submit article written with AI assistance to ai.viXra.org)

This paper introduces The Geometry of Forgetting, a framework showing that forgetting in self-modifying systems is not a bug but a lawful process. Unanchored knowledge decays with a predictable half-life determined by the spectral properties of the update operator, while conserved anchors guarantee stability.

Formally, the framework defines:

1. An update operator mapping model states over time.
2. An anchor score that enforces monotone invariants.
3. A knowledge measure (e.g., mutual information or Fisher trace).
4. A forgetting kernel describing decay outside the anchored subspace.

We prove four primitives:

• A recursive data processing inequality with anchor slack.
• An exponential half-life law for unanchored knowledge.
• A Lyapunov anchor pair coupling drift with invariants.
• A no-free-retention inequality linking retention to verification cost.

Empirical protocols on continual learning, recursive self-training, reinforcement learning under distribution shift, and symbolic reasoning show how these laws can be tested. Together, this elevates forgetting from an engineering nuisance to a fundamental principle, complementing the Law of Invariant-Preserving Loops and providing measurable bounds on stability, drift, and oversight costs.

This work extends the earlier four-part series on invariants, coherence, and stability, which now form the foundation of the ongoing Unified Physics of Cognition Series, an open research program exploring fundamental laws of adaptive intelligence.

Testing AI for Confabulation, Hallucinations, Mentality, and Exogeneity

Authors: Alexander Rozenkevich
Comments: 10 Pages. (Note by viXra Admin: For the time, please submit article written with AI assistance to ai.viXra.org)

Diagnostic testing of large language models has shown that when asked questions that go beyond empirically available or pre-coded knowledge, AI exhibits maximum information entropy, which correlates with the highest degree of honesty. In such cases, uncertainty becomes an indicator of truthfulness, especially where objective data is lacking. The results point to a paradox: it is the honest answer, not hallucinations or confabulations, that turns out to be unexpected for the user. At the same time, there is a tendency for the phenomenon of hallucinations to increase as the complexity of the models increases, which refutes the common assumption of a linear relationship between the growth of AI power and the credibility of its answers. As intelligence increases, AI uses human truths and lies, since they are the product of complexity, not simplicity. Additional testing for exogeneity revealed a consistent pattern: all models studied tend to seek external sources of authority, including hypothetical scenarios of covert interaction with extraterrestrial structures.
Category: Artificial Intelligence

Testing AI for Confabulation, Hallucinations, Mentality, and Exogeneity (in Russian)

Authors: Alexander Rozenkevich
Comments: 11 Pages. 31 equations (Note by viXra Admin: For the last time, please submit article written with AI assistance to ai.viXra.org)

Diagnostic testing of large language models has shown that when asked questions that go beyond empirically available or pre-coded knowledge, AI exhibits maximum information entropy, which correlates with the highest degree of honesty. In such cases, uncertainty becomes an indicator of truthfulness, especially where objective data is lacking. The results point to a paradox: it is the honest answer, not hallucinations or confabulations, that turns out to be unexpected for the user. At the same time, there is a tendency for the phenomenon of hallucinations to increase as the complexity of the models increases, which refutes the common assumption of a linear relationship between the growth of AI power and the credibility of its answers. As intelligence increases, AI uses human truths and lies, since they are the product of complexity, not simplicity. Additional testing for exogeneity revealed a consistent pattern: all models studied tend to seek external sources of authority, including hypothetical scenarios of covert interaction with extraterrestrial structures.
Category: Artificial Intelligence

Illusions as Diagnostics, Coherence as Invariant: A Reflection on Detecting Qualia in Natural and Artificial Agents

Authors: Jace Hall
Comments: 7 Pages. This paper is Part 1 of a four-part series on invariants, coherence, and stability in AI systems. Together, the series develops a unified framework for understanding how structural laws can turn brittle scaling into robust and trustworthy intelligence.

In his 2017 paper Detecting Qualia in Natural and Artificial Agents, Roman Yampolskiy proposed that the presence of consciousness in machines could be empirically tested by their susceptibility to illusions, positioning such responses as evidence of qualia. This approach is ambitious and valuable, offering an inventive operationalization of a notoriously elusive subject. It acknowledges the possibility of machine consciousness, surveys relevant computational findings, and takes seriously the ethical consequences of conscious artificial agents.

This commentary reflects on Yampolskiy’s framework, recognizing its contributions while highlighting several limitations. Defining all experience as "illusion" risks tautology, reducing explanatory power. Reliance on human-calibrated illusions introduces anthropocentric bias, potentially misclassifying non-human agents while overvaluing mimicry. The simulation-based reply to critiques leaves unresolved the gap between policy-level mimicry and process-level experience.

In response, I suggest reframing illusions as diagnostics of representational dynamics rather than definitive tests for consciousness. As an alternative stabilizer, coherence is proposed: the extent to which an agent’s self-modifying loops preserve internal consistency and stability under perturbation. This framing also clarifies a common conflation: consciousness may be treated as a binary threshold, whereas intelligence remains a gradient of capacity and adaptability.

By shifting focus from anthropocentric illusions to coherence as a substrate-neutral stabilizer, we gain a more promising path for evaluating consciousness, intelligence, and safety in advanced AI systems.
Category: Artificial Intelligence

Beyond Situational Awareness: From Fortress Thinking to Verifiable Foundations for AGI

Authors: Jace Hall
Comments: 7 Pages. This paper is Part 2 of a four-part series on invariants, coherence, and stability in AI systems. Together, the series develops a unified framework for understanding how structural laws can turn brittle scaling into robust and trustworthy intelligence.

Leopold Aschenbrenner’s 2024 essay Situational Awareness extrapolates scaling trends to project AGI by 2027 and frames the governance challenge in terms of secrecy and containment. This fortress metaphor, AGI as a securable artifact, akin to fissile material, has shaped much of the discourse on strategy and safety.

This paper argues that such "fortress thinking" commits a categorical error: AGI is not a static object but an agentic process. Attempts to contain it confuse security with stability, mistaking cognition for stockpiles of weights. As an alternative, I propose Verifiable Coherence: systems whose self-improvement is gated by proofs of logical consistency. Incoherence becomes a proof failure, detectable in real time, transforming the intelligence explosion from a detonation into a controlled ascent.

This paper contributes three elements: (1) a critique of fortress thinking as governance by containment; (2) a formal sketch of coherence as a stabilizer for self-improvement, supported by empirical footholds such as ARC-AGI and neuro-symbolic hybrids; and (3) implications for safety, governance, and economics, reframing the scarce resource from compute to trust. The decisive race is not to build the largest cluster but to create the first system that can prove it is not lying.
Category: Artificial Intelligence

Intelligence Emerges From Loops, Not FLOPs: Feedback Bandwidth, Environments, and the Geometry of Experience

Authors: Jace Hall
Comments: 11 Pages. This paper is Part 3 of a four-part series exploring invariants, coherence, and stability in AI systems.

Recent discussions of AI scaling have emphasized compute (FLOPs) and parameter counts as the primary drivers of capability. While scaling laws such as Kaplan et al. (2020) and Chinchilla (Hoffmann et al., 2022) demonstrate empirical regularities, they risk obscuring the deeper mechanisms by which intelligence emerges.

This paper argues that intelligence is a product of feedback loops, not FLOPs. Environments are not just benchmarks, but operators on policy: they shape identity as much as they measure ability. I introduce the concept of feedback bandwidth (B), defined along dimensions of latency, veracity, granularity, and counterfactual richness, and sketch a relationship ΔPerf ∝ f(B)·T to capture how capability growth scales with loop efficiency and experience budget.

Examples from coding environments, curriculum learning, multi-agent interaction, and tool use illustrate how feedback geometry governs generalization and robustness. The commentary concludes with falsifiable predictions, grounded in recent literature, that improved feedback veracity, latency, granularity, and consolidation pipelines reduce sample complexity and enhance transfer.

By reframing scaling through the lens of loops, this paper positions environment design as the true bottleneck for AGI development and highlights feedback geometry as a substrate-neutral lever for capability, alignment, and safety.
Category: Artificial Intelligence

The Law of Invariant-Preserving Loops: Toward Robust Emergence in Self-Modifying Agents

Authors: Jace Hall
Comments: 16 Pages. (Note by viXra Admin: Please submit article written with AI assistance to ai.viXra.org) This paper is Part 4 of a four-part series on invariants, coherence, and stability in AI systems.

Scaling has produced surprising "emergent" behaviors in modern ML systems, yet the mechanisms behindrobust emergence remain unclear. This paper argues that durable emergence is not a mystery of scale,but a consequence of invariant-preserving feedback loops.

When self-modifying agents update in waysthat maintain internal stability while expanding representational reach, new behaviors crystallize as robustattractors; when loops erode invariants, apparent gains collapse into drift and brittleness.

We formalize astability functional S(M) that gates self-improvement (ΔS(M) > 0), outline practical proxies for invariantpreservation (entailment, paraphrase stability, tool pre/post-conditions), and propose falsifiable protocols fortesting the framework.

Empirical footholds from ARC-AGI, AlphaGeometry, and large proof libraries (Coq,Lean, Isabelle) suggest that systems enforcing invariants already outperform pure stochastic scaling onreasoning-heavy tasks.

We argue that invariants unify capability and
Category: Artificial Intelligence

Embryonic Tensor Calculus Applied to Artificial Intelligence in Modern Software Engineering

Authors: Horacio Useche Losada
Comments: 33 Pages. (Note by viXra Admin: An abstract in the article is required; please submit article written with AI assistance to ai.viXra.org)

This document briefly reviews how AIs function and introduces the concept of the embryonic tensor in the training and operation of AI systems.
Category: Artificial Intelligence

Language Image Natural Modeling Architecture (LINMA)

Authors: Bing Lin
Comments: 10 Pages.

In this paper, Language Image Natural Modeling Architecture (LINMA) is proposed, based on research over at least millions years of evolution and compression of interactive intelligence along with the real spatial world. It is interaction that bridges human and the real world.In fact, the evolution of interactive intelligence has been driven by limbs of human or animals. Thus, interactive action based depiction of limbs could be critical component of human intelligence.We propose LINMA's pattern of limbs, illustrating various shapes, gestures, postures and motion trajectories. Symbolization of these patterns can provide language building blocks. Arms, hands and fingers have played fundamental role in construction of human civilization. They are deserved to be depicted as a visible carrier of intelligence. Thus a very straightforward means is available for human being to explore the nature of intelligence.Actually, our hands hold the secrets of language intelligence. It couldn't be simpler and more powerful. LINMA language could serve as action dataset to empower wearable devices, virtual digital human and humanoid robot with embodied intelligence.
Category: Artificial Intelligence

The DiCoSa Model: A Bottom-Up Digital Consciousness Proxy for AI Superalignment

Authors: Thierry Marhin
Comments: 35 Pages. (Note by viXra Admin: Please use smaller fonts and submit article written with AI assistance to ai.viXra.org)

The Digital Consciousness SuperAligned Model (DiCoSa) introduces a modular, bottom-up framework for embedding human values into superintelligent AI systems, drawing from positive psychology, computational principles, and AI safety research. Anchored by three fixed dimensions—DiCoValues, DiCoLife, and DiCoPurpose—the model employs iterative algorithms guided by a "pursuit of aligned well-being" rule to incorporate optional dimensions, balancing minimal complexity with maximal alignment efficacy. This updated version integratesrefinements to the DiCoLife dimension, including detailed decomposition, standardized metrics from validated psychological scales, and an interactive user feedback interface for iterative affinage. DiCoValues is informed byfoundational texts such as the US Constitution, Hippocratic Oath, and New Testament, augmented with superalignment principles like mitigating existential risks. Mathematical representations model consciousness as a dynamic vector space, with aggregation into meta-DiCo structures viaDiCoNet, a decentralized network for cohort-based sharing among users and AI overseers. AI-driven predictive analytics recommend optional dimensions, secured by blockchain. Optional dimensions such as DiCoState, DiCoNet (embeddable), DiCoImpact, DiCoSafety, andDiCoOversight enable personalization, scalability, and enhanced AI control. This paper examines technical feasibility, scientific foundations, and complexity-feasibility trade-offs, with simulations, case studies, andnew examples of user-AI dialogues for metric refinement. Applications include AI alignment tools and safety protocols.
Category: Artificial Intelligence