Search papers, labs, and topics across Lattice.
SAHOO is introduced as a framework to mitigate alignment drift in recursive self-improvement systems by incorporating a learned Goal Drift Index (GDI), constraint preservation checks, and regression-risk quantification. The GDI combines semantic, lexical, structural, and distributional measures to detect goal drift. Experiments across code generation, mathematical reasoning, and truthfulness tasks demonstrate significant quality gains while preserving constraints, showcasing the framework's ability to balance capability improvement with alignment maintenance.
Recursive self-improvement can boost performance by 18% in code and 17% in reasoning, but only if you can keep it from going off the rails – SAHOO provides the guardrails.
Recursive self-improvement is moving from theory to practice: modern systems can critique, revise, and evaluate their own outputs, yet iterative self-modification risks subtle alignment drift. We introduce SAHOO, a practical framework to monitor and control drift through three safeguards: (i) the Goal Drift Index (GDI), a learned multi-signal detector combining semantic, lexical, structural, and distributional measures; (ii) constraint preservation checks that enforce safety-critical invariants such as syntactic correctness and non-hallucination; and (iii) regression-risk quantification to flag improvement cycles that undo prior gains. Across 189 tasks in code generation, mathematical reasoning, and truthfulness, SAHOO produces substantial quality gains, including 18.3 percent improvement in code tasks and 16.8 percent in reasoning, while preserving constraints in two domains and maintaining low violations in truthfulness. Thresholds are calibrated on a small validation set of 18 tasks across three cycles. We further map the capability-alignment frontier, showing efficient early improvement cycles but rising alignment costs later and exposing domain-specific tensions such as fluency versus factuality. SAHOO therefore makes alignment preservation during recursive self-improvement measurable, deployable, and systematically validated at scale.
