Answer to 'How might a herd of interns help with AI or biosecurity research tasks/questions?'

Abstract

Discussion in an online forum engages with the topic of utilizing a large group of interns for research tasks, particularly in the areas of artificial intelligence (AI) or biosecurity. It evaluates the potential benefits and challenges of this approach. One concept proposes an “interpretability mine,” where a group of interns would analyze different parts of an AI model like GPT-2. The main challenges to this idea, the author argues, are developing a paradigm for interpretability work, establishing effective organizational structures, and creating suitable tools for the interns. Additionally, there’s contemplation about building a visualized reasoning model mapping out existential risk/recovery scenarios. The idea of interns prioritizing and reviewing long lists of project ideas is also considered. Ultimately, the feasibility of these ideas depends on various factors, including training and organizational capabilities. - AI-generated abstract.