recursive self-improvement (AI)
A scenario where AI systems improve their own capabilities, leading to a feedback loop where each improvement enables faster subsequent improvements
As defined in Understanding the Most Viral Chart in Artificial Intelligence | Odd Lots
Where it comes up · 1
Most important claims · 10 of 20Search all claims for “recursive self-improvement (AI)” →
- Time horizon charts plot the difficulty of tasks AI systems can complete over time, where difficulty is measured by how long it takes humans to complete those same tasks under identical conditions, showing an exponential increase in AI capabilities with doublings occurring roughly every 4 months in recent trends.Understanding the Most Viral Chart in Artificial Intelligence | Odd Lots
- Human baselines are established by recruiting talented humans with relevant expertise (e.g., software engineers for software tasks, ML engineers for ML tasks) who are not familiar with the specific task beforehand, timing how long it takes them to complete the task successfully using the same tools the AI will use, and averaging across approximately three baselines per task.Understanding the Most Viral Chart in Artificial Intelligence | Odd Lots
- Many people in the Bay Area who are concerned about AI safety have the intuition that the best way to influence AI's future is to work inside the industry building the technology, but this creates a structural problem: the logic of 'if I don't build it, someone else will' encourages continuous advancement, and combined with competition between labs and geopolitical competition with China, even safety-motivated developers face incentives to build more advanced systems.Understanding the Most Viral Chart in Artificial Intelligence | Odd Lots
- METR is a research nonprofit dedicated to measuring whether and when AI systems might pose catastrophic risks to humanity, specifically focusing on threats from AI autonomy by assessing how autonomous AI systems are and the scale, length, and difficulty of tasks they can perform independently.Understanding the Most Viral Chart in Artificial Intelligence | Odd Lots
- Chinese AI models are approximately 9-12 months behind US models by release timing, and the gap is potentially even larger when measured by time horizon, with some evidence suggesting Chinese models perform better on benchmarks than on held-out problems, possibly due to benchmark overfitting.Understanding the Most Viral Chart in Artificial Intelligence | Odd Lots
- Real-world software engineering tasks differ from METR benchmarks in several ways: they involve code quality concerns (elegant vs. messy code), collaboration with other engineers, larger codebases, adversarial scenarios where others modify code you're working on, and require verification overhead because humans must review AI work without context, leading to productivity gaps not captured by benchmark scores.Understanding the Most Viral Chart in Artificial Intelligence | Odd Lots
- Joel suspects that METR's task distribution is increasingly becoming a narrower slice of all possible tasks and specifically overlaps more with the exact task distributions used by AI labs for training, meaning METR is measuring progress on tasks optimized for (rather than independent test of) lab capabilities.Understanding the Most Viral Chart in Artificial Intelligence | Odd Lots
- There is a strange dynamic in the AI industry where lab CEOs and safety researchers make similar alarming claims about AI risks (destroying the world, taking over), creating a 'Baptist and bootlegger' alignment between people building the technology and people warning about it, which is unusual compared to most industries.Understanding the Most Viral Chart in Artificial Intelligence | Odd Lots
- METR's time horizon charts have become the industry standard benchmark for measuring AI progress, and investment decisions are explicitly being made based on these charts as investors interpret them as direct evidence of AI capability and business value.Understanding the Most Viral Chart in Artificial Intelligence | Odd Lots
- OpenAI and Anthropic were founded with exotic corporate structures (private company owned by nonprofit, etc.) because the founders took seriously the existential risks of AI and wanted to self-limit, suggesting at least some industry actors are not purely motivated by profit.Understanding the Most Viral Chart in Artificial Intelligence | Odd Lots