Towards Scientific Intelligence: A Survey of LLM-based Scientific Agents
What this work does
Organizes scientific agents around planners, memory, action spaces and verifiers, distinguishing prompt-native planning from learned planning.
Connection to discovery
A mechanism-level index for separating harness design, policy learning and evidence verification in a discovery agent.
Evidence
Inspected the architectural taxonomy and examples of reflective, tool-using and learned planners in the revised survey.
Scope and limitations
A narrative survey does not establish that one architecture causes better discovery; its benchmark coverage must be checked against original papers.
THE RESEARCH CONVERSATION
Discussion & reactions
Share a question, an insight, or a replication experience. Sign in with GitHub to join the conversation.