Explainer Videos

An Introduction to Cooperative MARL

Start here. It builds the subject from a single learning agent to cross-play evaluation, and each idea creates the problem that makes the next one necessary.

An Intro to Multi-Agent Reinforcement Learning in Cooperative Environments19:38Watch on YouTubeJoint actions, partial observability, shared rewards and credit assignment; CTDE, VDN and QMIX; learned communication under capacity, cost, delay and failure; partner dependence, ad hoc teamwork and cross-play. Ends on four questions the field has not settled.

LLMs as Cooperative Agents

The frontier chapter, in video form. It carries the same ideas to teams of language models and is careful about which claims current research supports.

When LLM Agents Work Together: Cooperative Multi-Agent Reinforcement Learning Meets Language Models13:32Watch on YouTubeDesigned versus learned collaboration; multi-LLM collaboration as a Dec-POMDP; why a shared reward over a joint response reopens credit assignment; MAGRPO and group-relative advantage; agent societies and what remains unsolved.

What the videos claim

Simplified environments shown in the videos are teaching models, not research benchmarks. In the frontier video, the sample return values on screen are illustrative rather than results from the cited papers, and every paper named is listed on the references page.

Watching alongside the written chapters

Video section	Written chapters
Joint actions, observations, shared reward	Background
Independent learning, CTDE, credit assignment, VDN, QMIX	Coordinate
Message capacity, cost, failure, learned protocols	Communicate
Partner dependence, diversity, ad hoc teamwork, cross-play	Adapt
Everything in the second video	LLMs as Cooperative Agents