What the F*ck Is Artificial General Intelligence?
4 hours ago
- #AGI
- #Machine Learning
- #Artificial Intelligence
- Artificial general intelligence (AGI) is a debated field, often seen as a Rorschach test due to hype and speculation.
- AGI is defined in terms of adaptation, likened to an artificial scientist, with foundational tools like search and approximation.
- Key architectures for adaptive systems include o3, AlphaGo, AERA, NARS, and Hyperon.
- Meta-approaches to AGI include scale-maxing (maximizing resources), simp-maxing (simplicity of form), and w-maxing (weakness of constraints).
- Examples of these approaches include AIXI, the free energy principle, and the expansion of language models ('The Embiggening').
- Current bottlenecks in AGI development are sample and energy efficiency, despite hardware improvements enabling progress.
- The conclusion suggests AGI will emerge from a fusion of tools and meta-approaches, with scale-maxed approximation currently dominant.