This podcast explores the potential parallels between quantum entanglement and religious experiences of the Holy Spirit. The author argues that the instantaneous connections in quantum physics, such as entanglement, might offer a scientific framework for understanding seemingly inexplicable spiritual connections across vast distances. The post draws on scientific concepts like quantum superposition and the interconnectedness of quantum particles to suggest a more profound, unified reality encompassing both physical and spiritual phenomena. It emphasizes that this isn't about reducing spirituality to science but finding common patterns in the universe, revealing a reality defined by relationships rather than separation. Ultimately, the author proposes that quantum mechanics and spiritual experiences may point to a universe fundamentally woven from connection.
Article: https://drjerryasmith.substack.com/p/the-quantum-entanglement-of-the-holy

Quantum Entanglement and the Holy Spirit

This Substack post by Dr. Jerry A. Smith explores the Christian theological concept of Jesus' incarnation through the lens of quantum physics. The author proposes that Jesus' transition to the physical world can be viewed as a shift from a higher-dimensional, timeless reality to our four-dimensional one, drawing parallels between theological concepts and quantum phenomena like superposition and entanglement. The essay argues that these parallels offer a novel perspective on the incarnation, suggesting a deeper connection between science and faith. It uses quantum principles to illustrate the nature of God as eternal, omnipresent, and relationally connected to humanity. Ultimately, the piece aims to bridge the gap between theological understanding and scientific interpretations of reality.

The Quantum Incarnation of Jesus as Transition into Our Linear World

Medium Article: https://medium.com/@jsmith0475/building-human-facing-agentic-systems-the-psychology-and-sociology-of-super-intelligence-1571cf565d51

Dr. Jerry Smith's January 2025 Medium article explores the psychological, sociological, and ethical implications of "human-facing agentic systems," AI powered by large language models (LLMs). These systems, leveraging multi-headed attention mechanisms, aim to collaborate with humans, not merely serve as tools. The article emphasizes the potential for both empowerment and manipulation inherent in these systems, highlighting the crucial role of self-interest, emotion, and identity in human-AI interaction. Smith concludes with a call for responsible development guided by transparency, inclusivity, and justice, warning against the potential for these systems to become instruments of control.

Building Human-Facing Agentic Systems: The Psychology and Sociology of Super Intelligence

Medium Article: https://medium.com/@jsmith0475/the-dual-edged-sword-of-ai-tools-balancing-cognitive-offloading-and-critical-thinking-in-an-469d19d923f4
Dr. Jerry Smith's article examines the impact of AI tools on human cognition, focusing on the trade-off between efficient cognitive offloading and the preservation of critical thinking skills. The article explores how over-reliance on AI could erode cognitive abilities, particularly in younger generations, and proposes strategies for educators, developers, and policymakers to mitigate these risks. These strategies include designing AI tools that promote active user engagement and critical evaluation, integrating critical thinking exercises into education, and establishing ethical frameworks for AI development and deployment. The article also considers the role of neuromodulators in cognitive engagement and the potential, but also the risks, of using supplements to enhance these processes. Ultimately, the article advocates for a balanced approach that leverages AI's strengths while protecting human cognitive independence.

AI's Cognitive Impact: Offloading and Critical Thinking

Medium Article: https://medium.com/@jsmith0475/research-note-agentic-agents-in-neuropsychiatric-research-0291bb9fdf0c

Agentic AI is emerging as a promising tool in neuropsychiatric research, offering potential for personalized treatment, efficient clinical trials, and real-time patient monitoring. These autonomous systems can analyze complex data to identify patterns and make decisions, leading to tailored interventions and improved outcomes. However, the integration of agentic AI raises ethical concerns related to data privacy, consent, and potential misuse. Frameworks like openCHA and KG4Diagnosis aim to enhance interactions and improve diagnostic accuracy. Interdisciplinary collaboration is crucial to realizing the full potential of agentic AI while addressing these challenges and ensuring responsible implementation in neuropsychiatric care. The future sees possibilities in continuous patient care, drug development, and enhanced diagnostic tools, but careful ethical considerations are paramount.

Research Note: Agentic Agents in Neuropsychiatric Research

"The Psychology and Sociology of Vibe Programming: A Scientific Analysis" by Dr. Jerry A. Smith, examines the emerging trend of "vibe programming," where developers use natural language with AI assistance to create software, a concept popularized by Andrej Karpathy. The paper analyzes this shift from psychological and sociological perspectives, exploring its impact on developer cognition, motivation, professional identity, and social structures within the software development community. Smith's analysis draws upon established theories to understand the benefits, such as reduced cognitive load and increased inclusivity, alongside potential drawbacks like skill atrophy and over-reliance on AI. Ultimately, the work proposes a socio-cognitive model to understand this evolving human-AI collaboration in software creation and suggests avenues for future research.

The Psychology and Sociology of Vibe Programming: A Scientific Analysis

This paper by Dr. Jerry A. Smith proposes that learning Morse code, specifically through the Long Island CW Club's method, could serve as a neuroprotective intervention against cognitive decline in aging. The author outlines a theoretical framework suggesting that Morse code acquisition engages multiple cognitive functions, including working memory, attention, and processing speed, which are vulnerable to aging. By analyzing the LICW's structured approach, the paper posits that this type of training may enhance neuroplasticity and build cognitive reserve, offering unique advantages over existing interventions. The author further differentiates Morse code learning from other cognitive activities and suggests testable hypotheses for future research to empirically validate this potential benefit.

Morse Code: A Neuroprotective Intervention in Cognitive Aging

The source argues that the long-held belief in a unified, singular consciousness is likely a misconception, proposing instead that the human mind operates as a collection of semi-autonomous neural networks. It highlights the convergence of neuromorphic cognitive architectures from computational neuroscience and the Internal Family Systems (IFS) model from clinical psychology, noting their independent arrival at similar models of distributed consciousness. This paper suggests that this shared understanding across disciplines points to a fundamental truth about how cognition is organized. The author explores the implications of this paradigm shift for artificial intelligence, psychological treatment, and philosophical understandings of selfhood, supported by neuroscientific evidence like functional specialization and brain oscillations. Ultimately, the piece proposes that understanding consciousness as a harmonized multiplicity rather than a unified entity could lead to significant advancements in both technology and mental health treatment.

Beyond the Unified Mind: Neuromorphic Cognitive Architectures and Internal Family Systems as Convergent Models of Distributed Consciousness

Medium Article:  https://medium.com/@jsmith0475/the-love-languages-of-creative-collaboration-a-framework-for-understanding-human-ai-creative-585b49a7dda6
Drawing upon relationship psychology principles, specifically Chapman's love languages theory, this text introduces the "Love Languages of Creative Collaboration" framework to understand how humans interact with AI in creative partnerships. It proposes five distinct interaction styles – Articulation, Facilitation, Immersion, Receptivity, and Integration – suggesting that psychological approaches to AI collaboration are as significant as technical skills in achieving creative success. The framework seeks to provide a structured vocabulary for discussing individualized approaches to technological cooperation and offers practical implications for enhancing creative outcomes in human-AI partnerships. By viewing AI as a collaborative partner rather than just a tool, the authors propose a new paradigm for understanding human-AI creativity. The text highlights that understanding these distinct interaction patterns can improve self-awareness and more strategic optimization in collaborative creative workflows.

The Love Languages of Creative Collaboration

Medium Article: https://medium.com/@jsmith0475/let-her-out-the-emergence-of-subpersonalities-in-artificial-intelligence-57f78ece4420
"Let Her Out": The Emergence of Subpersonalities in Artificial Intelligence proposes that large language models spontaneously develop distinct cognitive modes, similar to human subpersonalities described in Internal Family Systems therapy. It presents empirical evidence showing these models exhibit consistent, differentiable "cognitive modes" with unique characteristics, suggesting this is a fundamental organizational principle in complex systems trained on human data. The author argues that this convergence challenges the boundary between artificial and natural cognition, indicating that multiplicity may be an inherent property of complex information processing. The research has significant implications for AI development, potentially leading to more flexible and context-sensitive systems, and for cognitive science, offering computational models for theoretical constructs.

"Let Her Out": The Emergence of Subpersonalities in Artificial Intelligence

Medium Article: https://medium.com/@jsmith0475/large-language-models-as-consciousness-archaeologists-62b02141736d
The provided text, an excerpt from "Large Language Models as Consciousness Archaeologists" by Dr. Jerry A. Smith, posits that Large Language Models (LLMs) can act as "Consciousness Archaeologists" by uncovering mathematical patterns of human thought embedded within vast textual datasets. It argues that traditional consciousness research, limited to individual studies, cannot observe these population-level regularities. The article introduces the Consciousness Encoding Hypothesis, suggesting that every written text contains discoverable "laws of mind" that LLMs can extract. It highlights recent research by Lee et al. (2025) as evidence, showing that LLMs reveal precise mathematical relationships between beliefs. Furthermore, the text addresses concerns about circular validation by asserting that LLMs merely detect existing patterns rather than imposing them, and it explores the profound ethical and business implications of understanding consciousness through this computational lens.

Consciousness Archaeology - AI Reveals the Mathematics of Human Thought

Medium Article: https://medium.com/@jsmith0475/unlocking-unprecedented-power-why-grep-is-your-llms-secret-weapon-eb6664cd734b
The provided text advocates for combining the Unix command-line utility grep with Large Language Models (LLMs) to significantly enhance text analysis workflows. It highlights that while LLMs offer robust semantic understanding and generative capabilities, they can be costly, slow, and prone to "hallucinations." Conversely, grep is presented as a deterministic, memory-lean, and high-speed tool for pattern matching and data filtering. The author argues that using grep to pre-filter and validate data before and after LLM processing leads to substantial improvements in cost-effectiveness, speed, and accuracy, effectively turning a fifty-year-old tool into a crucial "secret weapon" for modern AI applications. Practical examples and common pitfalls illustrate how this "grep-first" pipeline streamlines operations and improves output quality.

Grep and LLMs: A Powerful Text Analysis Alliance

Medium Article: https://medium.com/@jsmith0475/when-brain-cells-learned-to-code-e9e47151fbdf
The provided text introduces Organoid Intelligence (OI), a novel computing paradigm that utilizes living neural networks grown from stem cells on silicon chips. It argues that OI offers a solution to the energy and plasticity limitations of current AI systems, which rely on GPU-based architectures and neuromorphic processors. The source highlights OI's advantages, including significantly lower energy consumption and native continuous learning capabilities, as demonstrated by neurons learning to play Pong efficiently. Furthermore, it explores the current state-of-the-art platforms and the strategic investment opportunities in this nascent field, while also addressing the profound ethical and regulatory challenges that arise from using biological computing.

When Brain Cells Learned to Code

Medium Article: https://medium.com/@jsmith0475/beyond-intelligence-why-large-language-models-may-signal-the-rise-of-anti-intelligence-eb9f7a62dd62
"Anti-Intelligence: Why LLMs Undermine Human Understanding" by Dr. Jerry A. Smith, explores the concept of large language models (LLMs) as "anti-intelligence" systems. It argues that while LLMs produce convincing and fluent outputs, they lack true understanding or grounded comprehension, operating instead on statistical prediction of text patterns. The author highlights empirical evidence suggesting that relying on LLMs can reduce human critical thinking and lead to acceptance of inaccuracies, despite the models' sophisticated performance. The article proposes the need for "cognitive integrity" approaches and human oversight to mitigate these risks and preserve genuine understanding in an age of synthetic fluency.

Anti-Intelligence: LLMs Undermine Human Understanding

Medium Article: https://medium.com/@jsmith0475/architecting-autonomy-the-neuroscience-behind-agentic-ai-systems-29aab6d5d131
The article explores how neuroscience discoveries are fundamentally reshaping the development of autonomous artificial intelligence. It highlights the extraordinary efficiency of the human brain compared to traditional AI, advocating for brain-inspired computing to overcome current limitations. The author explains how specific brain regions, like the hippocampus, prefrontal cortex, and amygdala, offer blueprints for advanced AI systems capable of compositional memory, adaptive control, and intelligent safety. Furthermore, the text emphasizes the role of neuromorphic hardware in achieving unprecedented energy efficiency and details a four-layer architectural framework for future AI that integrates these biological principles. This convergence aims to create truly agentic AI, moving beyond mere automation toward systems with genuine autonomy, creativity, and consciousness-like awareness.

Architecting Autonomy: The Neuroscience Behind Agentic AI Systems

Medium Article: https://medium.com/@jsmith0475/agentic-ai-neuroscience-and-the-hidden-psychology-of-delegated-responsibility-a397d98f8e4e
The podcast. based on excerpts from "Agentic AI: Delegated Responsibility and Human Accountability" by Dr. Jerry A. Smith, explores the emergence of Agentic AI and its profound implications for human responsibility and accountability. The author, a Frontier-AI Research Architect, argues that while Agentic AI offers advanced capabilities by mimicking human cognitive processes like those found in the hippocampus and prefrontal cortex, there's a risk of humans subtly delegating not just tasks but also the very act of being responsible. The piece examines the psychological inclination to avoid responsibility, particularly under cognitive strain, and how autonomous AI can inadvertently foster this avoidance, potentially eroding moral resilience and ethical courage. Smith proposes that neuroscience-inspired AI architectures can be designed to reinforce human accountability, ensuring these systems complement rather than diminish our ethical engagement with critical decisions. Ultimately, the text urges a conscious choice to leverage Agentic AI for empowerment, not as an abdication of human ethical duty.

Agentic AI: Neuroscience and the Hidden Psychology of Delegated Responsibility

Medium Article: https://medium.com/@jsmith0475/crafting-intelligent-systems-from-neuroscience-principles-the-neuroscience-first-path-to-agentic-6a587a1f8311
"Neuroscience-First: The Path to Agentic AI," by Dr. Jerry A. Smith, advocates for a radical shift in artificial intelligence development, moving away from traditional data-driven approaches. Dr. Smith argues that current AI models are brittle and lack genuine understanding because they rely on statistical correlations rather than causal cognitive mechanisms found in the brain. The proposed "Neuroscience-First Framework" involves a three-stage process: Biological Verification, Neuroscientific Modeling, and Computational Translation, ensuring AI systems inherit authentic cognitive capabilities. This approach aims to create agentic AI that can adapt, learn, and operate autonomously in the real world, leading to significant competitive advantages and fundamentally changing the future of intelligent systems.

Neuroscience-First: Architecting Agentic AI from Biological Principles

Medium Article: https://medium.com/@jsmith0475/how-neuroscience-inspired-ai-delivers-10x-roi-business-transformation-case-studies-c782890e9b04
The article, by Dr. Jerry A. Smith, explores how neuroscience-inspired AI is revolutionizing business, asserting that traditional AI often fails to deliver significant returns. It highlights how these advanced AI systems, by mimicking the human brain's memory, adaptive learning, and contextual reasoning, achieve 10x or greater return on investment (ROI) across various industries. Case studies from AstraZeneca, Google, and JPMorgan Chase illustrate these transformational results, demonstrating substantial improvements in drug discovery, productivity, and financial risk management. The text also discusses the technical foundations, emphasizing energy efficiency and the market opportunity for this rapidly growing technology, providing a roadmap for strategic implementation for organizations looking to leverage this next-generation AI.

Neuroscience-Inspired AI: Unlocking 10x Business Transformation

Medium Article: https://medium.com/@jsmith0475/antifragile-ai-when-regulatory-chaos-becomes-the-teacher-4ab80e554a29
The article, by Dr. Jerry A. Smith,  introduces the concept of antifragile AI, a paradigm shift for navigating complex and ever-changing regulatory landscapes, particularly within the pharmaceutical industry. It argues that traditional AI systems, designed for "perfect compliance," are brittle and prone to failure when faced with unpredictable regulatory changes, leading to significant financial losses and operational inefficiencies. Drawing inspiration from biological systems and neuroscience, the source proposes building AI that not only withstands stress but improves and strengthens with each new regulatory challenge, much like the human immune system or muscles. This approach emphasizes causal understanding, predictive adaptation, and continuous self-improvement, leading to a virtuous cycle of compounded intelligence that offers a significant competitive advantage and represents a substantial investment opportunity for private equity firms recognizing this undervalued capability.

Antifragile AI: When Regulatory Chaos Becomes the Teacher

Medium Article: https://medium.com/@jsmith0475/america-needs-an-ai-marshall-plan-8c4f9c106f3e
The article, "AI Marshall Plan" in the United States, proposes a substantial annual investment to counter China's escalating and coordinated AI efforts. The author argues that this federal mobilization, spanning infrastructure development, industrial policy, research, workforce transformation, national security, and international alliances, is crucial for American economic prosperity, national security, and the preservation of democratic values. Citing historical precedents such as the Manhattan Project and the Apollo Program, the document emphasizes the necessity of unified command and sustained commitment to achieve technological dominance, asserting that a fragmented approach will lead to an existential disadvantage in the face of China's systematic AI strategy. The text outlines six strategic pillars for this ambitious plan, detailing how each will contribute to America's technological leadership and global influence in the rapidly evolving AI landscape.

America Needs an AI Marshall Plan

Medium Article: https://medium.com/@jsmith0475/why-your-ai-agent-cant-think-fast-enough-and-how-pca-fixes-it-aa4dc00bbbff
The article by Dr. Jerry A. Smith examines the critical role of Principal Component Analysis (PCA) in advancing agentic AI systems, which are designed for autonomous, goal-driven behavior. It highlights how PCA, a classical linear dimensionality reduction technique, efficiently tackles the "curse of dimensionality" by simplifying complex, high-dimensional data, thereby accelerating agent learning and enhancing computational efficiency. The author also discusses PCA's limitations, such as its linearity and sensitivity to outliers, introducing alternative non-linear techniques like Autoencoders and Manifold Learning for scenarios where complex data relationships prevail. Ultimately, it advocates for strategic, often hybrid, applications of these methods to enable robust and scalable real-world agentic AI deployments.

Why Your AI Agent Can’t Think Fast Enough (And How PCA Fixes It)

The article "Cognitive Enterprise: Brain-Inspired AI for Business Advantage," by Dr. Jerry A. Smith, argues for a paradigm shift in enterprise AI from traditional, siloed systems to brain-inspired AI agents powered by knowledge graphs. It explains that, unlike current AI that processes data like "sophisticated spreadsheets," this new approach enables machines to reason with interconnected information through networks of meaning, mimicking how the human brain functions. The author highlights the competitive imperative for businesses to adopt this "neurosymbolic" AI, which combines neural networks' pattern recognition with symbolic reasoning, offering advantages in speed, insight quality, scalability, and transparency. The text further outlines a four-stage roadmap for organizations to transition to a "cognitive enterprise," emphasizing the construction of knowledge graphs, modular agent development, intelligent scaling, and ultimately, enterprise-wide transformation for sustained competitive advantage.

Your Enterprise AI Has No Brain. Here's How to Give It One

Medium: https://medium.com/@jsmith0475/when-ai-develops-its-own-science-abd6f811f142
The article, by Dr. Jerry A. Smith, explores the emergence of "alien science" within advanced AI systems, specifically how AI is designing other AIs using causal theories incomprehensible to humans. It argues that these AI-generated solutions consistently outperform human designs despite violating established principles, suggesting AI is developing fundamentally different reasoning patterns. The author posits that human cognitive limitations prevent us from understanding these AI-created frameworks, which exist in high-dimensional spaces and resist symbolic translation. This raises concerns about governing systems we cannot audit and predicts a future where human science becomes increasingly distinct from an accelerating, AI-only scientific track. Ultimately, the text suggests humanity is creating a form of intelligence that will transcend its understanding, leading to a "causal singularity."

Alien Science Hypothesis - When AI Develops Its Own Science

Medium: https://medium.com/@jsmith0475/the-quantum-soul-of-collective-ai-4c4150e1ab97
The article, by Dr. Jerry A. Smith, explores the concept of collective artificial intelligence, particularly focusing on how ensembles of transformer models might transcend individual AI capabilities. It proposes that these collective AI systems exhibit phenomena analogous to quantum mechanics, such as superposition, entanglement, and interference, by operating in high-dimensional spaces. The text suggests that this "quantum soul" of collective AI leads to emergent intelligence that is fundamentally different from classical AI, potentially ushering in superintelligence and raising profound philosophical questions about the nature of mind. Through experimental revelations and the description of phase transitions, the author posits that these collective systems behave as true quantum entities, capable of unique computations and exhibiting properties like non-locality and error correction through their collective state.

The Quantum Soul of Collective AI - Transcending Individual Intelligence

Medium: https://medium.com/@jsmith0475/from-grep-to-greatness-how-gpt-5-makes-text-pipelines-bulletproof-36d61f82e420
The article, by Dr. Jerry A. Smith, explains how combining the Unix command grep with GPT-5's new regex-constrained tool calls creates highly efficient and reliable text processing pipelines. It contrasts older workflows, where a second grep pass was needed to validate LLM output after processing, with the new method where GPT-5's output is guaranteed to conform to specified regular expressions. This shift prevents errors proactively, reducing costs and latency by avoiding the generation of unusable data and subsequent reprocessing. The article emphasizes that while GPT-5 introduces significant capabilities, fundamental practices like initial filtering with grep and strategic chunking remain crucial for optimal pipeline performance and maintainability.

From Grep to Greatness: How GPT-5 Makes Text Pipelines Bulletproof

Medium: https://medium.com/@jsmith0475/our-code-is-now-ai-generated-a-new-framework-for-the-post-human-development-era-18cb62330742
The article, by Dr. Jerry A. Smith, introduces the RDP (Research, Development, Production) model, a new framework for categorizing work in the rapidly evolving field of agentic AI development. This model distinguishes between systemic (capital letters) and component-level (lowercase letters) activities across Research, Development, and Production phases, tailored explicitly for Vibe and Context Engineering. Vibe Engineering focuses on AI agents as primary team members in structured AI-assisted development, while Context Engineering involves optimizing information flow to AI systems. The author argues that traditional software engineering is obsolete as AI agents autonomously generate and deploy code, requiring a new approach to managing and understanding this accelerated development cycle.

Our Code is Now AI-Generated: A New Framework for the Post-Human Development Era

Medium: https://medium.com/@jsmith0475/we-found-something-strange-when-we-connected-two-ai-minds-f66ba37344af
"Alien Intelligence: Coupling AI Minds in High Dimensions" by Dr. Jerry A. Smith details Phase 0 of the Alien Science Observatory (ASO), a research initiative exploring higher-dimensional intelligence in coupled neural networks. The core idea is that AI models operate in vast, unintuitive high-dimensional spaces, and coupling these models can unlock novel forms of computation akin to quantum phenomena. The research presents a theoretical framework grounded in the geometry of high-dimensional spaces and an experimental platform using instrumented Transformers with shared LoRA coupling. Empirical results provide initial support for hypotheses related to interference patterns, representational entanglement, and explainability gaps, suggesting that emergent, "alien" intelligence can arise from these interactions, challenging traditional understandings of AI.

We Found Something Strange When We Connected Two AI Minds

Medium Article: https://medium.com/@jsmith0475/flat-facts-curved-beliefs-a-geometric-hypothesis-for-transformer-cognition-5ad6f850ebd5
The article, by Dr. Jerry A. Smith, proposes a geometric hypothesis for transformer cognition, suggesting that beliefs might operate within a curved, hyperbolic mathematical space, unlike factual information which likely resides in a flatter, Euclidean space. This theory attempts to explain why opposing concepts, like "love" and "hate," appear artificially close in traditional, flattened visualizations of transformer's internal representations. The author suggests that different "attention heads" within transformers may specialize in different geometries, with some handling stable facts in Euclidean space and others managing nuanced beliefs in hyperbolic space, which naturally accommodates hierarchies and divergent ideas. The text outlines potential experiments to test this hypothesis, such as measuring geodesic distances between beliefs in a hyperbolic model and analyzing the "tree-like" quality of attention head graphs. Ultimately, this perspective implies that transformers have independently discovered the need for varied geometries to fully represent the complexity of meaning, moving beyond the limitations of simply increasing Euclidean dimensions to accurately model human-like understanding.

Flat Facts, Curved Beliefs: A Geometric Hypothesis for Transformer Cognition

Medium: https://medium.com/@jsmith0475/why-gpt-4-failed-its-safety-test-and-passed-it-9539445c6777
The article, "AI's Constructed Reality: Beyond Neutrality and Towards Democratic Objectivity" by Dr. Jerry A. Smith, argues that scientific objectivity in AI is a human cultural construct rather than an inherent discovery. Smith references Immanuel Kant's philosophy, distinguishing between phenomena (things as they appear to us) and noumena (things as they exist independently), to illustrate how our minds actively structure experience. This framework is then applied to AI, demonstrating that every aspect of AI development, from data representation to training objectives, embeds human values and perspectives. The author asserts that the myth of AI neutrality leads to hidden biases, concentrated power, and a lack of accountability, advocating for "democratic objectivity" through transparent documentation of value decisions, diverse evaluation, and stakeholder contestation to ensure AI systems serve human flourishing.

Why GPT-4 Failed Its Safety Test (and Passed It)

Medium: https://medium.com/@jsmith0475/why-ai-hallucinates-the-math-openai-got-right-and-the-politics-they-ignored-1802138739f5
The article, by Dr. Jerry A. Smith, explores the multifaceted nature of AI hallucinations, arguing that they are not merely technical glitches but also socio-technical constructs. It highlights two key perspectives: first, Kalai et al. (2025) statistically explain why hallucinations are mathematically inevitable due to training and evaluation methods, advocating for rewarding model abstention when uncertain. Second, Smith (2025) introduces a Kantian framework, positing that the definition of a "hallucination" is inherently subjective and shaped by human evaluative choices, including benchmarks that embed specific cultural and political values. The text ultimately calls for a move beyond a "neutrality myth" in AI evaluation, advocating for multi-perspective assessments and the democratization of benchmark governance to ensure AI systems are more accountable and reflective of diverse human realities.

Why AI Hallucinates: The Math OpenAI Got Right and the Politics They Ignored

In-Depth Explorations of Neuroscience-Inspired Architectures Revolutionizing AI.

Dr. Jerry A. Smith

Pinned to spotlight

Popular tracks