Liquid Reasoning Transformers: A Sudoku-Based Prototype for Chess-Scale Algorithmic Tasks
This work addresses the challenge of scalable algorithmic reasoning for tasks like chess, though it is incremental as it builds on transformer architectures with novel mechanisms.
The paper tackled the problem of structured reasoning in AI by introducing the Liquid Reasoning Transformer (LRT), which uses adaptive depths and iterative updates to correct errors, achieving 98.68% digit accuracy and 36.30% full-puzzle accuracy on Sudoku without symbolic rules or search.
The Liquid Reasoning Transformer (LRT) is a transformer architecture designed for inference with adaptive depths using iterative changes, discard-based correction, and a learned stopping mechanism. Instead of relying on a single feedforward pass, the model updates a recurrent reasoning token across multiple internal steps, allowing it to correct early errors and allocate computation based on input difficulty. We evaluate the LRT on Sudoku as a controlled testbed for structured reasoning and show that it achieves strong performance, reaching 98.68% digit accuracy and 36.30% full-puzzle accuracy without using symbolic rules or search. Analyzing internal patterns shows that the discard and stop gates play different, important roles in stabilizing inferences and adjusting computational depth. We discuss how these mechanisms extend naturally to chess-scale reasoning tasks and outline extensions for multi-token reasoning and larger domains.