Search papers, labs, and topics across Lattice.
This paper explores the concept of recursive self-design in AI systems, proposing a framework that emphasizes the modification of the design space itself through AI-assisted methods. By evaluating various public systems against four operational criteria, the authors highlight the Darwin Goedel Machine (DGM) as a key example, demonstrating significant performance improvements in benchmark tasks after iterative self-modification. The findings underscore the potential for AI systems to not only evolve their capabilities but also to refine their own development processes through structured feedback mechanisms.
Recursive self-design enables AI systems to enhance their own architecture and performance, achieving up to 50% improvement in key benchmarks through iterative self-modification.
Recursive self-design refers to AI-assisted modification of the mechanisms by which an AI system is built, evaluated, and improved. This paper treats MetaAI not as a mature paradigm, but as a working term for a human-seeded, AI-expanded development pattern in which the design space itself becomes a target of modification. We propose an operational evidence framework with four criteria: inspectable target system, meta-level modifier, feedback-directed selection, and recursive continuation. We then map public systems, including Darwin Goedel Machine (DGM), STOP, Goedel Agent, and ShinkaEvolve, against these criteria. DGM provides the most direct currently reported evidence: its published results show improvement from 20% to 50% on SWE-bench Verified and from 14.2% to 30.7% on full Polyglot after 80 iterations, with ablations suggesting that both open-ended exploration and self-improvement contribute. Finally, we provide MetaAI-Mini, a reproducible HumanEval-based protocol and codebase. Because no completed model run is included in this build, MetaAI-Mini is reported as a protocol rather than as an experimental result.
