SAGE: New Framework for Scalable AI Governance in Large-Scale Search Systems
Researchers have developed SAGE, a framework that uses human judgment calibrated with AI to evaluate relevance in large-scale search systems like LinkedIn Search. The system combines natural-language policies, precedent examples, and an LLM judge in a feedback loop to create consistent, scalable evaluation signals. The approach achieved a 0.25% lift in LinkedIn daily active users while reducing evaluation costs by 92× through model distillation.
SAGE addresses a fundamental challenge in AI systems: evaluating relevance at scale while maintaining human oversight quality. Traditional approaches rely on engagement metrics or sparse manual review, which often miss high-impact failures. The framework operationalizes human product judgment through a bidirectional calibration loop where natural-language policies, curated precedents, and an LLM surrogate judge co-evolve to resolve semantic ambiguities and transform subjective relevance judgment into a multi-dimensional, executable rubric with near human-level agreement. To make this practical at industrial scale, the researchers applied teacher-student distillation to transfer high-fidelity judgments into compact models at 92× lower cost. When deployed in LinkedIn Search, SAGE enabled policy-driven model iteration, detected regressions invisible to engagement metrics, and contributed to a measurable 0.25% lift in daily active users.
What's missing
The study does not discuss potential limitations of the approach, such as how the framework handles novel or edge-case scenarios not covered by existing precedents, computational overhead of the calibration loop itself, or generalizability to search domains beyond LinkedIn.
What different sources said
- arXiv cs.AICenter
SAGE: Scalable AI Governance & Evaluation
Related
Genetic Drift, Not Selection, Drives Rapid Feather Color Evolution in Island Bird Radiation
A new study of an island bird radiation found that rapid evolution of feather coloration is driven primarily by genetic drift in small populations rather than sexual or ecological selection. The research integrated whole-genome data with detailed plumage measurements across complete species sampling to test whether signaling trait evolution correlates with speciation rates. The findings suggest that neutral demographic processes play a central role in generating phenotypic diversity during island radiations, challenging assumptions about the mechanisms driving rapid evolution.
New AI Model Improves Prediction of Therapeutic Peptide Function from Protein Sequences
Researchers developed a lightweight CNN classifier that predicts whether peptide sequences have therapeutic properties, trained on a database of 54,655 peptides across 48 functional categories. The model uses a novel negative sampling strategy to reduce false positive rates from over 60% in previous approaches to 2.1%. This advancement could accelerate drug discovery by enabling faster computational screening of peptide candidates before expensive experimental testing.
Study Shows Different Metabolic Stress Models Produce Distinct Effects on Human Neuronal Networks
Researchers tested three common in vitro metabolic stress models on human-derived neuronal networks and found each produced different patterns of neuronal activity and cell damage. The models tested were hypoxia alone, oxygen-glucose deprivation (OGD), and hypoxia combined with glutamate exposure. The findings suggest that choice of experimental model significantly affects results and that combining electrophysiological and structural analyses is important for accurately assessing metabolic stress in stroke research.