How To Fail Until You Succeed
A viral guide applies startup iteration principles to AI safety, arguing most projects fail due to slow execution.
A viral post on the AI forum LessWrong, titled 'How To Fail Until You Succeed' by Luc Brinkman, argues that most AI safety projects fail because they move too slowly and get stuck on initial ideas. Drawing directly from startup methodology, the post contends that successful founders succeed not by getting their first idea right, but by being 'relentlessly resourceful,' iterating constantly, and pivoting when needed. The guide is part of a five-part sequence on launching impactful AI safety work and applies core entrepreneurial concepts like execution, adoption, and the 'ship fast, kill fast' mentality to the unique challenges of reducing existential risk.
The post breaks down the path to impact with the formula Impact = Adoption x Effectiveness. It stresses that competent execution alone is insufficient; builders must first 'fall in love with the problem' of existential risk, identify a high-impact sub-problem, and only then devise solutions. To achieve adoption, the author recommends techniques from books like 'The Lean Startup' and 'The Mom Test,' such as talking to users and testing the riskiest assumptions. The core advice is to optimize for learning over pride, build scrappy prototypes quickly, and be willing to discard what doesn't work to find a solution that gains real traction.
- Applies startup 'ship fast, kill fast' iteration to AI safety, arguing most projects fail from slow execution and attachment to initial ideas.
- Proposes the impact formula 'Impact = Adoption x Effectiveness' and advises falling in love with the problem of existential risk first.
- Recommends resources like 'The Lean Startup' and 'The Mom Test' for de-risking projects through user feedback and rapid prototyping.
Why It Matters
Provides a concrete, entrepreneurial framework for AI researchers and builders to increase their projects' real-world impact and adoption.