A lantern-maker intern points at an intermediate concat layer and says, “That is the agent.” What is the correct distinction?
Select an answer to reveal the explanation.
Short Explanation
A fusion layer is a meeting hall for features, not a stage manager. Agents plan tools and roles; fusion just joins representations.
Full Explanation
Fusion modules combine modality representations at chosen depths. Agent orchestrators plan or sequence tools and roles. Treating a concat or cross-attention block as an agent collapses that distinction. Fusion type (early/intermediate/late) does not make the module an agent.