KUNGFU.AI Presents: Are Emergent Abilities of LLMs a Mirage?
"Are Emergent Abilities of LLMs a Mirage?" explores the concept of emergence in AI, particularly in Large Language Models (LLMs). Emergence refers to complex behaviors that arise unexpectedly as systems grow in scale. The presentation discusses whether the novel abilities seen in LLMs, such as Few-Shot Learning and logical reasoning, are genuine or merely statistical phenomena. Real-world and simulated examples of emergent behaviors, like traffic flow and bird flocking, are used to illustrate the concept.
The presentation also addresses the unpredictability and limitations of LLMs, citing critiques from figures like Noam Chomsky and rebuttals from others like Scott Aaronson. The session aims to provoke thought on the implications of emergent behaviors in AI.
This was originally presented at SXSW 2024 and was represented to our team.