Linus's 20% is not a false positive rate -- how has nobody caught this? Let's assume that "the 0.1% false positive rate [Google] tried to make such a big deal about last week" is actually a false positive rate by the normal definition.
The false positive rate is defined as (the number of false positives over (false positives + true negatives)). However, Linus quite obviously calculated the 20% as (false positives over positives), i.e. (200 / 1000). If Linus happened to have 200,000 true negatives (i.e., non-spam messages that were not flagged as spam), a number that I'm making up because he did not disclose one, then his false positive rate would be (200 / 200,200) ~= 0.1%.
Think about it... whether his email address was harvested by all the spambots in the world or none at all would have no effect on the fact that out of 200,200 legit messages, 200 were incorrectly flagged. This is why the false positive formula doesn't even include true positives! The 800 true positives (actual spam messages) don't matter to the formula. Therefore, neither does the 1000 total (true+false) positives. Don't divide by it.
The false positive rate is defined as (the number of false positives over (false positives + true negatives)). However, Linus quite obviously calculated the 20% as (false positives over positives), i.e. (200 / 1000). If Linus happened to have 200,000 true negatives (i.e., non-spam messages that were not flagged as spam), a number that I'm making up because he did not disclose one, then his false positive rate would be (200 / 200,200) ~= 0.1%.
Think about it... whether his email address was harvested by all the spambots in the world or none at all would have no effect on the fact that out of 200,200 legit messages, 200 were incorrectly flagged. This is why the false positive formula doesn't even include true positives! The 800 true positives (actual spam messages) don't matter to the formula. Therefore, neither does the 1000 total (true+false) positives. Don't divide by it.