Data Dowsing: Determining Data Collection Priorities

Ryan Marinelli

Keywords: LLM, model saturation, foundation models

TL;DR: Data Dowsing is a method to determine which content is most needed to improve foundation models.

Abstract:

This work proposes a novel framework, data dowsing to determine which data is needed to improve LLMs.
This framework is based on estimating influence and imposing simplifications based on concept domains
to circumvention computational intractability.

Serve As Reviewer: ~Christian_Salomonsen1

Submission Number: 27