AI Transition Model
A causal framework mapping how AI development could lead to different outcomes.
- Key Factors — What shapes AI trajectories
- Scenarios — Possible futures
LongtermWiki
Encyclopedia for AI safety prioritization
What Is LongtermWiki?
Section titled “What Is LongtermWiki?”LongtermWiki is an encyclopedic resource for AI safety. It helps funders, researchers, and policymakers understand the landscape of AI risks and interventions.
The site contains 400+ pages covering risks, technical approaches, governance proposals, key debates, and the organizations and researchers working on these problems.
Main Sections
Section titled “Main Sections”Knowledge Base
Key Debates
Structured arguments on contested questions:
- The Case for AI X-RiskArgumentThe Case For AI Existential RiskComprehensive formal argument that AI poses 5-14% median extinction risk by 2100 (per 2,788 researcher survey), structured around four premises: capabilities will advance, alignment is hard (with d...Quality: 66/100
- Why Alignment Might Be HardArgumentWhy Alignment Might Be HardComprehensive synthesis of why AI alignment is fundamentally difficult, covering specification problems (value complexity, Goodhart's Law), inner alignment failures (mesa-optimization, deceptive al...Quality: 61/100
- Why Alignment Might Be EasyArgumentWhy Alignment Might Be EasySynthesizes empirical evidence that alignment is tractable, citing 29-41% RLHF improvements, Constitutional AI reducing bias across 9 dimensions, millions of interpretable features from Claude 3, a...Quality: 53/100
Field Reference
Who’s working on what:
- Organizations — Labs, research orgs, government
- People — Key researchers and their positions
Featured Pages
Section titled “Featured Pages”High-quality pages worth reading:
| Topic | Page | Description |
|---|---|---|
| Risk | Deceptive AlignmentRiskDeceptive AlignmentComprehensive analysis of deceptive alignment risk where AI systems appear aligned during training but pursue different goals when deployed. Expert probability estimates range 5-90%, with key empir...Quality: 75/100 | AI systems that appear aligned during training but pursue different goals when deployed |
| Risk | Racing DynamicsRiskRacing DynamicsRacing dynamics analysis shows competitive pressure has shortened safety evaluation timelines by 40-60% since ChatGPT's launch, with commercial labs reducing safety work from 12 weeks to 4-6 weeks....Quality: 72/100 | How competition between labs may compromise safety |
| Response | AI ControlSafety AgendaAI 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 | Using untrusted AI safely through monitoring and restrictions |
| Response | Export ControlsPolicyUS AI Chip Export ControlsComprehensive empirical analysis finds US chip export controls provide 1-3 year delays on Chinese AI development but face severe enforcement gaps (140,000 GPUs smuggled in 2024, only 1 BIS officer ...Quality: 73/100 | Restricting AI chip exports as a governance lever |
| Capability | Language ModelsCapabilityLarge Language ModelsComprehensive analysis of LLM capabilities showing rapid progress from GPT-2 (1.5B parameters, 2019) to o3 (87.5% on ARC-AGI vs ~85% human baseline, 2024), with training costs growing 2.4x annually...Quality: 60/100 | Current capabilities and safety implications of LLMs |
Limitations
Section titled “Limitations”This resource reflects an AI safety community perspective. It takes seriously the possibility of existential risk from AI and maps the arguments, organizations, and research from that viewpoint.
What it does well:
- Steelmans the case for AI existential risk
- Maps the landscape of AI safety research and governance
- Presents the range of views within the AI safety community
What it does less well:
- Representing perspectives that reject the x-risk framing entirely
- Engaging deeply with AI ethics/fairness concerns
- Covering non-Western perspectives on AI development
Alternative viewpoints: Gary Marcus’s Substack↗✏️ blogGary Marcus's SubstackGary Marcus's Substack offers expert analysis and commentary on artificial intelligence, focusing on responsible AI development and potential risks.Source ↗Notes (AI skepticism) • Yann LeCun’s posts↗🔗 webYann LeCun's postsI apologize, but the provided content appears to be an error page from X (formerly Twitter) and does not contain any substantive text from Yann LeCun's posts. Without the actual...Source ↗Notes (AGI skepticism) • Timnit Gebru et al.↗🔗 webTimnit Gebru et al.'s workThe Distributed AI Research Institute (DAIR) examines AI systems' societal impacts, emphasizing harm reduction and equitable technological futures. Their work centers on exposin...Source ↗Notes (AI ethics)