Search papers, labs, and topics across Lattice.
The Mendel G\"odel Machine (MGM) introduces innovative self-modification strategies for coding agents by leveraging comparative signals from multiple task trajectories and reference agents. By employing reaction-norm mutation and cross-lineage hybridization, MGM enhances the self-improvement process beyond traditional single-trajectory methods. Experiments demonstrate that these strategies lead to faster convergence and improved performance on coding tasks, as evidenced by results on SWE-bench and Polyglot benchmarks.
Self-improving coding agents can achieve unprecedented efficiency and generalizability by learning from multiple task trajectories simultaneously.
Self-improving coding agents that iteratively rewrite their own source code have demonstrated impressive performance on coding tasks. However, existing solutions generally derive self-modification from a single failure trajectory at a time, overlooking rich comparative signals available in the agent's expanding archive of past attempts. According to Mendelian principles of controlled inheritance, we introduce Mendel G\"odel Machine (MGM). In addition to the general single-trajectory clonal mutation, MGM includes two new types of self-modification that better utilizes evidences accumulated: the reaction-norm mutation edits an agent based on its trajectories on multiple tasks simultaneously, and the cross-lineage hybridization edits an agent using the trajectory of a reference agent from another lineage on the same task. Under an additive fitness landscape model, we prove theoretically and demonstrate via controlled surrogate simulation that the new strategies facilitate a faster and better convergence over single-trajectory baselines. Experiments on SWE-bench and Polyglot confirm MGM's consistent improvement in performance, efficiency, and generalizability.
