Sensi: Learn One Thing at a Time -- Curriculum-Based Test-Time Learning for LLM Game Agents
This addresses the challenge of sample inefficiency in LLM agents for game-playing, though it is incremental as it builds on existing test-time learning approaches.
The paper tackles the problem of LLM agents requiring many interactions to learn tasks in unknown environments by introducing Sensi, a curriculum-based test-time learning architecture for game agents, which achieved 50-94x greater sample efficiency than comparable systems, solving 2 game levels with one version and completing its learning curriculum in 32 attempts with another.
Large language model (LLM) agents deployed in unknown environments must learn task structure at test time, but current approaches require thousands of interactions to form useful hypotheses. We present Sensi, an LLM agent architecture for the ARC-AGI-3 game-playing challenge that introduces structured test-time learning through three mechanisms: (1) a two-player architecture separating perception from action, (2) a curriculum-based learning system managed by an external state machine, and (3) a database-as-control-plane that makes the agents context window programmatically steerable. We further introduce an LLM-as-judge component with dynamically generated evaluation rubrics to determine when the agent has learned enough about one topic to advance to the next. We report results across two iterations: Sensi v1 solves 2 game levels using the two-player architecture alone, while Sensi v2 adds curriculum learning and solves 0 levels - but completes its entire learning curriculum in approximately 32 action attempts, achieving 50-94x greater sample efficiency than comparable systems that require 1600-3000 attempts. We precisely diagnose the failure mode as a self-consistent hallucination cascade originating in the perception layer, demonstrating that the architectural bottleneck has shifted from learning efficiency to perceptual grounding - a more tractable problem.