Keen On America Episode 2534: Why Generative AI is a Technological Dead End
May 15, 2025
Peter Voss, CEO of Aigo.ai and a pioneer in AI who coined 'Artificial General Intelligence' in 2001, critiques generative AI as a misguided venture. He argues that large language models (LLMs) are fundamentally flawed due to their lack of memory and inability to learn incrementally, calling them a technological dead end. Voss warns of an impending bubble burst in the industry, drawing parallels to past economic manias. He advocates for a return to foundational principles in AI development to truly advance towards human-like intelligence.
AI Snips
Chapters
Books
Transcript
Episode notes
LLMs Are A Technological Dead End
- Large language models (LLMs) are spectacular in tasks like translation and summarization but cannot lead to Artificial General Intelligence (AGI).
- LLMs lack incremental learning ability and thus represent a technological dead end for AGI.
LLMs Cannot Learn Incrementally
- LLMs need all training data upfront and cannot update their models incrementally in real time due to backpropagation constraints.
- This structural limitation causes models to be read-only and requires expensive retraining for updates.
LLM Hallucinations Are Inherent
- Hallucinations are intrinsic to LLMs due to their statistical nature and are unlikely to disappear as models scale.
- Once committed to an answer, LLMs generate justifications that may be fabricated to maintain conversation coherence.
