The Problem: Curiosity Fatigue
The default behavior of Large Language Models — delivering complete, authoritative answers on demand — poses a structural challenge to education. When students receive fully synthesized responses to complex questions, the cognitive labor required for comprehension is outsourced to the model.
Over time, this produces curiosity fatigue: a measurable decline in the willingness to engage in effortful inquiry when a frictionless alternative is available.
The Solution: The Curiosity Gap
Spark is an open-source library of steering contexts — plain-text instructional layers applied at the system-prompt level — that reconfigure LLM behavior for classroom use.
Instead of providing 100% of the information, Spark operates on an 80/20 Discovery Model: it offers approximately 80% of the necessary context, intentionally leaving a Curiosity Gap. This requires the student to "close the loop" by providing the final synthesis, identifying a flaw, or predicting a result.
See Spark in Action
Experience the difference between a default LLM and a Spark-steered AI across Science, Math, and History.