Korva is a scroll of the internet’s best ideas.
At CERN, the particle physics lab, two quite isolated teams work on the same particle accelerator at different places on it. They don't share their results until they publish. Over time the organization learns what works best in each team, and neither team gets stuck. Academia works this way, and free markets work extremely well because of competition.
Software now fits the same pattern, because engineering is less of the bottleneck and the real question is what the right thing to build is. Two people at one organization recently built separate versions of the same software, one for the US and one for Japan, as two different Lovable projects instead of forking each other's code. In the old way of building software, there would have been one track, and it would have turned into Franken software trying to fit the needs of both groups.
A good way to combine them is to open a new project, have it check out the other one, and bring over the three things worth keeping. Then run a split test or experiment to see whether the change improves the metrics for the customers being served.
What is a local minimum?
Picture a landscape of hills and valleys where the goal is to reach the lowest point. A search that only moves downhill from where it stands will settle in the nearest valley, even when a deeper one lies beyond the next ridge. That nearest valley is a local minimum. Computer scientists use the term in optimization, and a standard fix is starting several searches from different points.
How do CERN's two detector teams work?
The Large Hadron Collider has two general-purpose detectors, ATLAS and CMS, built and run by separate collaborations. Each analyzes its own collision data and keeps its work independent until it is ready to publish. Both teams announced the discovery of the Higgs boson on July 4, 2012. Two independent confirmations reached the same result.
Why do split tests settle product debates?
A split test shows two versions of a product to randomly chosen groups of users and compares a chosen metric. Random assignment means differences between the groups can be credited to the change itself. The method descends from the controlled experiments the statistician Ronald Fisher designed for agricultural research in the 1920s. Websites today run such tests routinely on small details like button wording.
More from All-In with Chamath, Jason, Sacks & Friedberg
