The Abstract

The Abstract@theabstract

0 followers
Follow

2025 episodes (2)

Beyond Aggregate - Using Information Theory to Forge True Teamwork in Multi-Agent LLMs
Ep. 02

Beyond Aggregate - Using Information Theory to Forge True Teamwork in Multi-Agent LLMs

When does a group of AI agents stop being a simple “collection” and start acting like a true “team”? We talk a lot about “collective intelligence” in AI, but how can we even measure it? In this segment, we dive into a new paper that tackles this question head-on. Using a clever guessing game where AI agents can’t communicate, researchers test what it takes for them to spontaneously coordinate. We discuss their new framework for measuring synergy and the surprising discovery about what kind of prompting - like invoking a ‘Theory of Mind’ - is needed to create a team that’s truly greater than the sum of its parts.

Defeating Brevity Bias and Context Collapse
Ep. 01

Defeating Brevity Bias and Context Collapse

How do we make AI smarter without constantly retraining it? This episode explores a new framework called ACE, or Agentic Context Engineering, that treats an AI’s instructions not as a static prompt, but as a dynamic, evolving “playbook.” We discuss how this method helps AI learn from its mistakes, accumulate knowledge, and even allows smaller models to compete with giants like GPT-4. Join us as we break down the problems of “brevity bias” and “context collapse” and explore a future of more efficient, self-improving AI systems.