Every card shows its question first. Reveal one at a time to test yourself, or
reveal all of them to use the deck as a revision sheet.
5 cards
Frontier
How can an LLM be viewed as an agent in an interactive system?
Frontier
An LLM can be viewed as a policy that receives information, such as a prompt or interaction history, and produces an action, such as a natural-language response or tool action.
Frontier
What might correspond to the observation, policy, action, and reward for an LLM agent?
Frontier
Observation: prompt, local context, messages, or interaction history
Policy: the LLM
Action: generated response or tool action
Reward: a task or collaboration score
The exact mapping depends on the system being studied.
Distinction
What is the difference between orchestrated LLM collaboration and learned collaboration?
Distinction
In orchestrated collaboration, designers specify roles, communication patterns, or interaction workflows.
In learned collaboration, agent behaviour is optimized toward a collaborative objective through learning.
Frontier
Why do coordination, communication, and credit assignment remain relevant when several LLM agents share one objective?
Frontier
Their outputs may need to fit together, agents may possess different information, and a shared outcome may not reveal which agent or interaction contributed most to success.
Frontier
What new research questions arise when LLM agents must cooperate with unfamiliar agents or larger agent societies?
Frontier
Important questions include:
how cooperation transfers to new models or partners;
how agents develop and share conventions;
how communication should scale;
how shared resources or objectives should be managed;
how collective behaviour should be evaluated as agent populations grow.