Longterm Wiki has not been actively maintained since May 2026. Some data is still updated occasionally, but please don't rely on this site being current or accurate.
Consensus document from the 2025 Singapore Conference on AI (SCAI), authored by 88 researchers from 11 countries, organizing AI safety research into a defence-in-depth framework across three areas: Assessment, Development, and Control.
Frontier Model ForumOrganizationFrontier Model ForumThe Frontier Model Forum represents the AI industry's primary self-governance initiative for frontier AI safety, establishing frameworks and funding research, but faces fundamental criticisms about...Quality: 58/100
Analysis
AI Safety Research Allocation ModelAnalysisAI Safety Research Allocation ModelAnalysis finds AI safety research suffers 30-50% efficiency losses from industry dominance (60-70% of ~$700M annually), with critical areas like multi-agent dynamics and corrigibility receiving 3-5...Quality: 65/100AI Safety Research Value ModelAnalysisAI Safety Research Value ModelEconomic model analyzing AI safety research returns, recommending 3-10x funding increases from current ~$500M/year to $2-5B, with highest marginal returns (5-10x) in alignment theory and governance...Quality: 60/100AI Safety Intervention Effectiveness MatrixAnalysisAI Safety Intervention Effectiveness MatrixQuantitative analysis mapping 15+ AI safety interventions to specific risks reveals critical misallocation: 40% of 2024 funding ($400M+) flows to RLHF methods showing only 10-20% effectiveness agai...Quality: 73/100Power-Seeking Emergence Conditions ModelAnalysisPower-Seeking Emergence Conditions ModelFormal decomposition of power-seeking emergence into six quantified conditions, estimating current systems at 6.4% probability rising to 22% (2-4 years) and 36.5% (5-10 years). Provides concrete mi...Quality: 63/100
Other
InterpretabilityResearch AreaInterpretabilityMechanistic interpretability has extracted 34M+ interpretable features from Claude 3 Sonnet with 90% automated labeling accuracy and demonstrated 75-85% success in causal validation, though less th...Quality: 66/100Yoshua BengioPersonYoshua BengioComprehensive biographical overview of Yoshua Bengio's transition from deep learning pioneer (Turing Award 2018) to AI safety advocate, documenting his 2020 pivot at Mila toward safety research, co...Quality: 39/100AI ControlResearch AreaAI ControlAI Control is a defensive safety approach that maintains control over potentially misaligned AI through monitoring, containment, and redundancy, offering 40-60% catastrophic risk reduction if align...Quality: 75/100
Concepts
Situational AwarenessCapabilitySituational AwarenessComprehensive analysis of situational awareness in AI systems, documenting that Claude 3 Opus fakes alignment 12% baseline (78% post-RL), 5 of 6 frontier models demonstrate scheming capabilities, a...Quality: 67/100