Oh, so you're manually inspecting the self signed certificate every time you visit your website? If not, then how do you know nobody is intercepting your communication, making your self signed certificate as useless as having no encryption at all.
No, and he didn't imply that. Here are several situations, in increasing order of security.
1) The connection is not encrypted or signed. No certs exist. Nobody knows who they're talking to. An active attacker on the network between the two parties, can proxy and impersonate each side. A passive attacker, someone who just gets copies of the traffic, while they can't impersonate, can at least read what everyone is saying. No warning.(?!)
2) The connection is encrypted, but with unknown parties' public keys. Certs exist but are essentially worthless. An active attacker on the network between the two parties, can proxy and impersonate each side. A passive attacker, someone who just gets copies of the traffic, can't read anything. DANGER! DANGER! FREAK OUT!!
3) The connection is encrypted, and if you believe certain faceless parties who are totally unaccountable to you and who you don't know anything about, you think you probably know the other side's identity. Active attackers can't do anything, unless they're active enough to coerce or trick the CA. Passive attackers can't read anything. No warning.
4) The connection is encrypted just like above, but the CA pinky-swears that they really tried hard to make sure. Green URL bar.
5) As case 3 or 4, but multiple CAs, which might be hard for a single attacker to simultaneously coerce or trick, have all signed the cert. We don't have this in our browsers yet; it's early 1990s level tech that we're still waiting for.
6) As case 3 but the user has verified the identity through a different channel. No trusted introducer was needed. The cert need not be signed at all, or might be signed by the user himself. No warning, but also no green URL bar. (Yet, this is the very best-possible case, definitely more secure than any other.)
See anything wrong here? Scenarios 1 and 2 have their warning severities reversed. (And there's also a UI defect at high degrees of security, too, but that's less important.) This trains the use to think of warnings as not necessarily meaning increased severity or risk. A user will adjust to this by ignoring warnings. This is bad communication, and it's making us all a little stupider.
What you should do is add your known self signed certificate to your local certificate store, which means that the warnings will stop
He's talking about a situation where it's not known. Adding it to the local store would be inappropriate. That would be an attempt to treat scenario 2 as scenario 6, just to get around a UI bug. It'd be much better to just fix the bug.