In-context learning as a new kind of symbolic-AutoML: Lyapunov conjecture for CoTs
Résumé
This short critique demonstrates how using Chain-of-though (CoT), i.e., In-context learning (ICL) can be considered as a symbolic-AutoML tool. Using ICL Pre-trained Large Language Models (PLMs/LLMs) can be directed to make data analysis, generate predictions or code for very specialised tasks without retraining.This is akin to doing meta modelling in AutoML, that pushing analysis into meta model doesn't remove the human analyst or reasoner. This is a common generalisation fallacy in many AI systems. Humans are actually are still in the loop in designing CoT datasets, i.e., set of facts and reasons. e provide basic definitions and identify what constitues as an AutoML task in ICL and usage of approach as symbolic-AutoML tool. Lyapunov-Conjecture for CoT is described illustrating limitations of the approach.
Origine | Fichiers produits par l'(les) auteur(s) |
---|