The article has a side-note saying: > “two models trained on the same data but with different algorithms will behave much more similarly than two models traine

The article has a side-note saying: > “two models trained on the same data but with different algorithms will behave much more similarly than two models trained on different data with the same algorithm. That’s because the training data contains the patterns in user behavior that algorithms mine and use when recommending new content.” Do we know how ‘true’ this statement is? Or how much % an algorithm must deviate to cause a % change in results? What about difference in data? If someone can point to somewhere to read up on this, I'd like to know!