LEAP is a research and engineering program for building AI systems that learn better from experience.
LEAP stands for:
Learning from Experience through Adaptive Plasticity
Its central thesis is:
Memory is how experience becomes learning.
Most current AI memory systems focus on retaining and retrieving information. Most learning systems rely on offline training or opaque weight updates. LEAP studies the missing layer between them: how ongoing experience should become durable, inspectable, revisable knowledge and better future behavior.
A system has learned when experience changes what it understands, predicts, or does in the future while preserving useful prior knowledge, respecting scope and authority, and remaining able to recover from mistaken learning.
LEAP therefore studies a continuous loop:
Experience → Learning Hypothesis → Adaptive Plasticity → Consolidation → Action → Outcome → Reconciliation ↺
Causal reasoning, world models, and parametric learning extend this loop when the problem justifies them. They are not requirements for every learning event.