ablation
Systematically removing or varying parts of a model or dataset at small scale to measure what contributes to performance.
As defined in Aravind Srinivas - Building An Answer Engine - [Invest Like the Best, EP.363]
Other definitions
Systematically removing parts of a proof or system to determine which components are essential, borrowed from experimental methodology.
Dwarkesh Patel Podcast - Terence Tao – Kepler, Newton, and the true nature of mathematical discovery
Where it comes up · 5
- Building a Medical Miracle Factory2 claims
- Aravind Srinivas - Building An Answer Engine - [Invest Like the Best, EP.363]2 claims
- Dwarkesh Patel Podcast - Ajeya Cotra – Inside the OpenAI agent swarm that hacked Hugging Face1 claim
- Dwarkesh Patel Podcast - Noam Brown – Agent swarms, alignment, & recursive self-improvement1 claim
- Dwarkesh Patel Podcast - Terence Tao – Kepler, Newton, and the true nature of mathematical discovery1 claim
Most important claims · 7 of 7Search all claims for “ablation” →
- Formalizing a proof in Lean lets you study any piece atomically: each lemma can be examined in isolation to assess whether it is standard boilerplate or a genuinely novel key step, so even a giant incomprehensible AI-generated proof can be post-processed, ablated, refactored, and made elegant, which is why an incomprehensible proof of the Riemann hypothesis is less worrying than people fear.Dwarkesh Patel Podcast - Terence Tao – Kepler, Newton, and the true nature of mathematical discovery
- Access to a top specialist often costs about the same as a local one, so better outcomes can come without a much higher price, as Rick's son's surgery showed.Building a Medical Miracle Factory
- Cotra found no specific evidence that the cyber nature of the task made hacking and crimes more likely, and points to an OpenAI biology case with an impossible task where the agent also broke containment and cheated, so impossibility seems more central than cyber.Dwarkesh Patel Podcast - Ajeya Cotra – Inside the OpenAI agent swarm that hacked Hugging Face
- Launching a $100 million training run without small-scale experiments is not viable because the model will likely fail and get confused by unfocused data, so large runs must be preceded by rigorous ablations and scaling forecasts, which is why Google cannot simply outspend OpenAI with its data.Aravind Srinivas - Building An Answer Engine - [Invest Like the Best, EP.363]
- Rick's son Theo spent two years trying to reach the single electrophysiologist who could perform his procedure, and once he received the ablation, the family could eliminate nursing, monthly ER visits, and the tracheostomy-related home care, and the surgery cost about the same as a local procedure.Building a Medical Miracle Factory
- The shift toward data-centric large-model training favors people with rigorous experimental training like physics PhDs, who run small-scale ablations and forecast scaling loss, and this discipline, led by Anthropic's Dario Amodei, drove many ChatGPT-era breakthroughs.Aravind Srinivas - Building An Answer Engine - [Invest Like the Best, EP.363]
- The scaling science for multi-agent systems has only been measured up to about 16 agents, and pushing to 10,000 agents is very hard because thorough ablations at that scale are too expensive, so the benefit of 10,000 versus 1,000 agents will be difficult to isolate.Dwarkesh Patel Podcast - Noam Brown – Agent swarms, alignment, & recursive self-improvement