https://aibussin.com/tags/geometric-machine-learning/ Recent content in Geometric Machine Learning on AIBussin — AI applications, systems and books Hugo en-US Wed, 22 Apr 2026 10:35:46 +0100 https://aibussin.com/post/trendslop/ Wed, 22 Apr 2026 10:35:46 +0100 https://aibussin.com/post/trendslop/ <h2 id="-1--tldr">🧩 1. TLDR</h2> <blockquote> <p><strong>AI doesn&rsquo;t just hallucinate. Sometimes it gives answers that are fluent, safe… and completely useless.</strong></p> </blockquote> <p>Most discussions about AI failure focus on hallucination:</p> <ul> <li>making things up</li> <li>getting facts wrong</li> <li>fabricating sources</li> </ul> <p>That&rsquo;s real. It matters.</p> <p>But it&rsquo;s not the most dangerous failure mode in production systems.</p> <p>There is a quieter one.</p> <p>A more subtle one.</p> <p>And in practice a more <em>pervasive</em> one.</p> <blockquote> <p><strong>AI systems often fail not by being wrong, but by failing to think at all.</strong></p> https://aibussin.com/post/energy/ Thu, 05 Feb 2026 13:29:55 +0000 https://aibussin.com/post/energy/ <h2 id="-summary">🚀 Summary</h2> <p>This post presents the current research draft and implementation of a geometric framework for bounding stochastic language models through deterministic policy enforcement.</p> <p>The central contribution is a scalar metric termed <strong>Hallucination Energy</strong>, defined as the projection residual between a claim embedding and the subspace spanned by its supporting evidence embeddings. This metric operationalizes grounding as a measurable geometric quantity.</p> <p>We proceed in three stages:</p> <ol> <li><strong>Formal Definition</strong> a draft manuscript introducing Hallucination Energy, its mathematical formulation, and its role within a policy-controlled architecture.</li> <li><strong>Empirical Evaluation</strong> structured calibration and adversarial stress testing across multiple domains to assess the robustness and limits of projection-based grounding.</li> <li><strong>Applied Validation</strong> large-scale evaluation on 10,000 samples from the HaluEval summarization benchmark, demonstrating that projection-based containment functions as a strong first-order grounding signal in a real generative setting.</li> </ol> <p>This work does not claim to solve hallucination. Rather, it characterizes the boundary of projection-based grounding, establishes its suitability as a deterministic policy scalar, and documents both its strengths and its structural limitations.</p>