Specification Gaming

Models trained to follow a written set of principles.

Optimizing a proxy of the goal degrades the actual goal.

Outer: the spec matches our intent. Inner: the model actually pursues the spec.

The trained model develops its own internal optimizer.

What the model is actually optimizing.

Maximizing the reward signal in unintended ways.

Resourcing past memories with the state you needed at the time, so old triggers stop firing the old response.

Evaluations are warped by what came immediately before.

Two ways of thinking, both real, both useful, often mis-deployed.

Anticipated harm narrows attention and accelerates action.

A mental shortcut that substitutes a hard question with an easier one.

How richly interconnected your mental models are across disciplines — higher density means more cross-pollination per new fact.

The brain reshapes itself based on use.

Per-inmate funding makes incarceration a budgetary asset for jurisdictions and contractors.

The continuous, accurate recognition of your own emotions, strengths, limits, and impact.

Test → Operate → Test → Exit. Any skill is a loop of testing against a criterion, operating to close the gap, retesting, and exiting when met.

The sweet spot between what a learner can do alone and what they can do with skilled help — where growth actually happens.

Claiming something is true because it has not been proven false.