Same Hardware, Different Settings: Why AI Models Believe Their Lies

Last episode explored how your brain deliberately corrupts its own memories. Today, Joey and the show pull on that thread—and uncover why the entire AI field has been using the wrong word all along. Drawing on neuroscience research and the U.S. government's own standards (NIST), we explore confabulation: the confidence trick your brain and AI models share. It's not hallucination. It's something deeper—where the line between productive dreaming and pathological memory fabrication runs through the same hardware, just different temperature settings. Key timestamps: 00:00:00 - Your brain corrupts its own memories (why last episode matters today) 00:01:31 - Why NIST and neuroscientists say confabulation beats hallucination 00:04:52 - Two types of confabulation in neuropsychology 00:10:05 - Sleep and hidden patterns: why dreaming is productive 00:11:53 - Same hardware, different settings 00:13:47 - How the human brain runs its reality filter 00:15:25 - The genuine belief in false memory (Rashomon witness) 00:18:46 - Temperature as the core metaphor --- Sources & further reading: • Armin Schnider — "Spontaneous confabulation and the adaptation of thought to ongoing reality": https://www.nature.com/articles/nrn1179 • Erik Hoel — "The overfitted brain: Dreams evolved to assist generalization": https://www.cell.com/patterns/fulltext/S2666-3899(21)00064-7 • Bender, Gebru, McMillan-Major, Mitchell — "On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?" (2021) — (originally published at FAccT 2021) • NIST — "AI 600-1: Generative AI Risk Profile" (July 2024): https://www.libertify.com/interactive-library/nist-ai-600-1-generative-ai-profile/ • Nora Benavidez and Giancarlo Mori — "Hallucination or Confabulation? Neuroanatomy as metaphor in Large Language Models": https://pmc.ncbi.nlm.nih.gov/articles/PMC10619792/ • Integrative Psych — "Hallucination vs. Confabulation: Rethinking AI Error Terminology": https://www.integrative-psych.org/resources/confabulation-not-hallucination-ai-errors • Wolters Kluwer — "Hallucination vs. Confabulation: Why the difference matters in healthcare AI": https://www.wolterskluwer.com/en/expert-insights/hallucination-vs-confabulation-why-difference-matters-in-healthcare-ai • Stanford HAI — AI Index 2026, hallucination measurement across 26 frontier models — (referenced via multiple aggregators) • Vectara — HHEM hallucination benchmark data (May/September 2026): https://www.elitecontentmarketer.com/ai-hallucination-statistics/ • Scientific American — "Why Lawyers Keep Citing Fake Cases Invented by AI": https://www.scientificamerican.com/article/why-lawyers-keep-citing-fake-cases-invented-by-ai/ • GC AI — "AI Hallucination Legal Cases: Sanctions Tracker": https://gc.ai/blog/ai-hallucination-legal-cases • Srivastava, Hinton, Krizhevsky, Sutskever, Salakhutdinov — "Dropout: A Simple Way to Prevent Neural Networks from Overfitting" (2014): https://www.jmlr.org/papers/v15/srivastava14a.html • Wagner, Gais, Haider, Verleger, Born — "Sleep inspires insight" (Nature, 2004) • biorxiv preprint — "Retrieval uncertainty drives rapid consolidation of confabulated memories in retrosplenial cortex" (2026): https://www.biorxiv.org/content/10.64898/2026.09.01.748171v1.full • Springer — "What is the Creative Value of AI Hallucinations?" (2026): https://link.springer.com/article/10.1007/s10796-026-10747-x • The Innovation: AI Plus — "Extracting creativity from hallucination: rethinking large language models" (2026): https://www.the-innovation.org/ai-plus/article/doi/10.59717/ipj.aiplus.2026.100008 • …and 1 more in the episode research notes This podcast episode was fully generated by AI — research, script, voices, and production. Built with Claude, Piper TTS, and automated pipeline tooling.