Sometimes I see quite clear that they (often the men, I keep a closer watch to their competitions) begin on inside edge and just before take-off they slip over to outside edge, and it's a Lutz!
I am very far from being an expert on these matters (maybe GKelly will help out!

.) But I think the idea is something like this.
Most defintions and descriptions of the proper technique for a flip jump go something like this one.
http://www.sk8stuff.com/f_recog/recog_j_flip.htm
The flip jump is a toe-assisted jump which has the same take-off and landing edges as the Salchow. It launches off a LBI edge (right toe-pick assist) and lands on the traditional RBO. The normal preparation is quite a bit different however. It is usually entered from a very straight (flat) forward glide on the left foot. The free foot (right) will be held up in the air in front of the skater. Just prior to the jump, the free leg will drop back, tap the ice and push the skater around in a LFO 3-turn, so that he is on a shallow LBI edge. The right foot goes directly behind the skater, picks the ice and “pulls” the skater up into the air.
Some skaters will substitute an RFI-LBI mohawk for the LFO 3-turn prior to the jump.
So there is quite a bit more to the description of the flip jump than just the take-off edge. If a skater does most of those things, but messes up slightly on some part of it (for instance, he might land on the wrong edge), it is still scored as a flip.
If he does all those things except that at the last moment of the take-off he slips over to the outside edge, then according to ISU protocol it is
still scored as a flip (not a Lutz), but the skater gets an "e" call and an automatic deduction in GOE. This is what is called informally a "lip."
Many people think this is wrong. They contend that the
entire definition of the flip jump is the take-off edge, and nothing else conntributes to the definition. So in this case, the jump should be scored as a Lutz (with no penalty) rather than as a faulty flip (with a penalty on GOE).
This seems to be a minority view in ISU circles. I guess it's like the definition of a horse. If you look up "horse" in the dictionary it says, "a four legged animal that ..." and then goes on to list some other features.
Now suppose a horse had one leg amptutated. So now it is not a "four-legged animal..."
But it is still a horse and not, for instance, a foot-stool, even though the definition of "foot-stool might be, "a three-legged...."
Same thing with a flip. If it looks like a flip and acts like a flip and quacks like a flip -- then it's a flip, despite the wrong take-off edge. (So says the ISU in its wisdom. Others disagree.)
Eevun said:
(just watched Kristoffer Berntsson's FS from CoC and he didn't receive an edge call for that) Because I haven't seen a skater skating on the outside edge all the way until the take-off.
Many wrong edge take-offs are not called. Its up to the Technical Specialist to call what he sees, either an "e" (extreme wrong edge) or a "!" (mild wrong edge), or sometimes no call because the Tech Specialist just didn't see it that way.
About the Lutz, I have seen skaters hold a long outside edge all the way down the ice. But it is a very straight and shallow one, so it is hard to be certain.
To me (an amateur observer, to say the least), the thing that distinguishes a Lutz most dramatically from other jumps is the counterrotation -- the blade is curving one way on the ice, but the skter somehow wrenches back the other way with the toe pick and with the upper body. That's what makes it look so cool, in my humble opinion.
Here is a great one. Even without playing the video, you can see in the still picture, by her shoulders and arms, that she is about to launch into the air spinning counterclockwise even though her skate is curving clockwise. The Lutz is the only jump that is like that.
http://www.sk8stuff.com/f_recog/recog_j_lutz.htm
If a skater cheats and slips over to the inside edge at the last moment, again, this is scored as a Lutz despite the wrong edge and should get an edge call and penalty in GOE. The protocols will show 3Lz e (this is a flutz).
(I apologize for being so long-winded if you already knew all that. Anyway, the argument is between those who think that the take-off edge is the
entire definition of the flip and Lutz, and those who think that there is more to it than that.)