English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

In this work we create agents that can perform well beyond a single,individual task, that exhibit much wider generalisation of behaviour to amassive, rich space of challenges. We define a universe of tasks within anenvironment domain and demonstrate the ability to train agents that aregenerally capable across this vast space and beyond. The environment isnatively multi-agent, spanning the continuum of competitive, cooperative, andindependent games, which are situated within procedurally generated physical 3Dworlds. The resulting space is exceptionally diverse in terms of the challengesposed to agents, and as such, even measuring the learning progress of an agentis an open research problem. We propose an iterative notion of improvementbetween successive generations of agents, rather than seeking to maximise asingular objective, allowing us to quantify progress despite tasks beingincomparable in terms of achievable rewards. We show that through constructingan open-ended learning process, which dynamically changes the training taskdistributions and training objectives such that the agent never stops learning,we achieve consistent learning of new behaviours. The resulting agent is ableto score reward in every one of our humanly solvable evaluation levels, withbehaviour generalising to many held-out points in the universe of tasks.Examples of this zero-shot generalisation include good performance on Hide andSeek, Capture the Flag, and Tag. Through analysis and hand-authored probe taskswe characterise the behaviour of our agent, and find interesting emergentheuristic behaviours such as trial-and-error experimentation, simple tool use,option switching, and cooperation. Finally, we demonstrate that the generalcapabilities of this agent could unlock larger scale transfer of behaviourthrough cheap finetuning.

Open-Ended Learning Leads to Generally Capable Agents