Will recursive self-improvement be observed in AI by 2030?

This market asks whether AI systems will achieve full recursive self-improvement (RSI) by end of 2030 — defined as autonomous performance of the complete research, ideation, implementation, and validation cycle to create more capable versions of themselves without human intervention in the core R&D process. Recent research shows impressive progress: the Darwin Gödel Machine improved its SWE-bench performance from 20% to 50% through iterative self-modification, SICA achieved 17-53% gains on coding benchmarks, and AlphaEvolve discovered the first improvement to Strassen's matrix multiplication algorithm in 56 years while also optimizing its own training infrastructure. A new startup, Recursive Superintelligence, raised $650M specifically to build recursively self-improving AI. However, these systems remain narrow — improving performance on specific benchmarks rather than achieving general intelligence improvements through full autonomous R&D. OpenAI's o3/o4-mini still rates below the "High" threshold for self-improvement capability. External sentiment on related questions sits around 29%. The quote reflects significant recent progress and increased investment in RSI research, balanced against the high bar set by the market's definition requiring demonstrated full autonomous R&D cycles for general capability improvements.