Episode

Ep.192: Expert Argues Large Language Models Will Never Achieve True Intelligence

Podcast
The Power AI Podcast
Published
Dec 18, 2025
Duration seconds
794
Processing state
not_requested
Canonical source
https://unboxedai.blogspot.com/p/podcasts.html
Audio
https://dts.podtrac.com/redirect.mp3/api.spreaker.com/download/episode/68916866/expert_argues_large_language_models_will_never_achieve_true_intelligence.mp3
JSON
/v1/public/podcasts/the-power-ai-podcast-7037522/episodes/ep-192-expert-argues-large-language-models-will-never-achieve-true-intelligence
Markdown
/podcast/the-power-ai-podcast-7037522/ep-192-expert-argues-large-language-models-will-never-achieve-true-intelligence.md

Actions

  • POST https://stenobird.com/v1/public/podcasts/the-power-ai-podcast-7037522/episodes/ep-192-expert-argues-large-language-models-will-never-achieve-true-intelligence/transcription-requests
    Idempotently request low-priority transcript generation for this episode.
  • GET https://stenobird.com/podcast/the-power-ai-podcast-7037522/ep-192-expert-argues-large-language-models-will-never-achieve-true-intelligence.md
    Read the agent-friendly Markdown representation of this episode resource.

Summary

Todays deep dive presents a significant argument challenging the notion that Large Language Models (LLMs) are close to achieving Artificial General Intelligence (AGI) by focusing on the language-intelligence fallacy. Benjamin Riley, a prominent voice in the field, asserts that LLMs are simply sophisticated emulators of communication and lack genuine thought or reasoning capabilities, a view supported by neuroscience indicating that language processing is separate from core cognitive functions. This critique suggests that scaling LLMs will not solve their inherent architectural limitations, leading Riley to liken them to "dead-metaphor machines" that are perpetually confined to their training data. Furthermore, other leading AI figures like Yann LeCun express skepticism about current methods, advocating instead for "world models" that learn from diverse physical data. Research confirms these limitations, concluding that these probabilistic systems cannot generate truly novel outputs, reaching an inescapable ceiling on creativity that restricts them to remixing existing knowledge.