Yep. I missed an invitation to a Google-hosted event at a conference I attended because the email (from an @google.com address, no less) got caught in Gmail's spam filter.
This is a problem with ML approaches, right? Instead of water boiling at "100C" it boils at "99.98C +- 0.04C". Normally this is ok, but sometimes it isn't!
I imagine most humans have error rates worse than that. And what does 'error rate' even mean in that context? A small delay or a catastrophic failure ending in death and destruction?
I would think that would be quite a good error rate. Especially when considering people in the hospital are often not in good health, possibly making their veins more difficult to find.
That is assuming by error you mean missing the vein. If error is defined as a fatal complication, then 1/1000 is terrifying.
If there are 50,000 people and 50,000 people experience problems, that's bad.
If there are 5 million people and 50,000 experience problems, that's fine?
Isaac Asimov's comments about world population increase involved something about this; the more people there are, the more each individual is dehumanised and rendered irrelevant (my paraphrasing).
No I don't think it's fine at all. I think Google, twatter, Facebook, et al don't care because 50 people and who they represent don't matter to them compared to the money they make.
When all rounded up it isn't even a single penny on the balance sheet. The owners of these businesses literally never even know from the their only view into the companies.
I have no idea why I'm being downvoted on this. Hackers can't do math or what?
I think your position is a little unrealistic. 50 people experiencing problems out of what, a billion? is pretty good. Do you think that if those billion people were served by 20 million small business e-mail providers, that none of those 20 million e-mail providers would ever make a mistake and affect their 50 customers?
Yup, I've had the same thing. Just kinda amusing and ironic since I didn't happen to care about that event but it makes one nervous about relying on spam filtering.
On the other hand, once you train it a bit, it is mostly remarkable good. For me, switching from fastmail.fm (which was pretty good itself) to Gmail gave me a big improvement in spam control.
Super curious about this response saying on gmail you have more control, because from where I am the “mark as spam” button does nothing but move things to the spam folder. In theory it should learn from that but when someone used my email address to sign up for AT&T no amount of marking things as spam will stop their emails landing in my inbox.
As in signed up for AT&T service? If so it's because it's not spam - it's misdirected mail, but there are tens of thousands of other Gmail users who think that messages almost identical to those are things they absolutely want to receive.
My point is that absent information that Google simply does not have no matter how creepy they get, there's literally no way they can identify such messages as spam - exactly the opposite in fact because probably 99.999% of such messages that they process are explicitly not spam.
The only way Google would have to identify that this message was not for you would be to get the subscriber information from AT&T and cross-reference it with name and address information they had for you - and even then most of the time they'd probably be wrong (e.g. if the email is coming to you but the account is actually in a family member's name).