Literacy perspective
How tokens, attention, parameters, sampling, variability, and model behaviour affect predictability.
What lecturers should notice
Separate observable behaviour from speculation about hidden internals, then manipulate generation conditions and repeat runs.
Related units
The role may be primary, secondary, supporting, integrated, or part of the attack surface; consult each unit’s metadata for the exact relationship.
Participants learn to stop treating model variability as inexplicable noise. They learn to hold a task stable, vary one generation condition, repeat trials, and use the resulting evidence to decide when variability is useful, when it is risky, and which controls are appropriate.
Participants learn that asking for improvement is not yet a reflection process. They learn to separate construction, judging, revision, and verification; define explicit conditions of satisfaction; and test whether critique produces evidence-backed improvement rather than confident churn.