He's right that data volumes have non-linear risk profiles.
He's wrong that there is evidence more data isn't better. While there are indications of this for advertising, it is definitely not the case for financial data.
And the other subtlety is that lots of low quality data is indeed useless, but small sums of high value data can do a lot. That high value data is what people are looking to steal. Having a little bit of user financial traction data, for example, is incredibly powerful. Much more so than, say, cross-website shared cookies or Amazon referral patterns.
And there is a whole class of data that has value proportional to the total sum of it you possess. A good example of that is surveillance data. Ubiquitous video coverage of an environment is much more useful (to both machines and humans) than partial coverage.
He's wrong that there is evidence more data isn't better. While there are indications of this for advertising, it is definitely not the case for financial data.
And the other subtlety is that lots of low quality data is indeed useless, but small sums of high value data can do a lot. That high value data is what people are looking to steal. Having a little bit of user financial traction data, for example, is incredibly powerful. Much more so than, say, cross-website shared cookies or Amazon referral patterns.
And there is a whole class of data that has value proportional to the total sum of it you possess. A good example of that is surveillance data. Ubiquitous video coverage of an environment is much more useful (to both machines and humans) than partial coverage.