In plain words
Artificial general intelligence (AGI) refers to AI that can handle a wide range of cognitive tasks rather than a narrow specialty. Definitions differ over the breadth and level of performance required.
A closer look
Generality concerns how many kinds of tasks a system can handle; performance concerns how well it handles them. Excelling at one benchmark does not establish broad competence. Assessing AGI therefore requires evidence across varied tasks, including unfamiliar problems.
AGI describes capability, not necessarily consciousness or independent action. Superintelligence refers to abilities beyond human levels; the singularity is a proposed broader transformation. These ideas should not be treated as interchangeable.
In practice
Imagine one system learning an unfamiliar design tool, solving a new mathematical problem, and planning a research project with broadly human-level competence. That illustrates the intended breadth of AGI, rather than proving it through any single demonstration.
A useful distinction
AGI has no single universally accepted threshold. A fluent chatbot or a high test score alone is not sufficient evidence.