AI Signal 538 2 feeds carried it
Scott Aaronson argues LLM intelligence emerged without explicitly engineered self-referentiality
↗
Scott Aaronson argues that LLMs achieved high-level intelligence without explicitly engineered self-referentiality, challenging the view that "strange loops" are a prerequisite for AI.
This challenges foundational theories of AI, particularly Douglas Hofstadter's view that intelligence requires "strange loops." It suggests that prediction and compression, rather than self-reference, are the core drivers of the intelligence observed in current LLMs.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
Scott Aaronson argues that LLMs achieved intelligence without explicitly building in self-referentiality or "strange loops."
LLMs' ability to discuss self-reference emerged as a free byproduct of universality and general pretraining, similar to how formal systems acquire self-referential capabilities.
The ideas vindicated by LLMs relate to prediction, compression, and Kolmogorov complexity rather than self-referentiality, although consciousness remains unexplained.
THE CLUSTER