I don't really agree with some of the content in the link you have provided.
Personas (are not superman)
Personas are not fictional per-se. A good persona will be heavily based on empirical research though their bio and photo may be fictional. This is to conceal the identity of the research participants. Regardless, the fictional part is intended to make the character believable - not as a key design guide. Having said that, the so called 'fictional' details are very often a mere twist on real biographic data.
To quote Alan Cooper (from About Face 3, chapter 5 - Modelling Users: Personas and Goals):
Personas are not real people, but they are based on the behaviours and motivations of real people we have observed and represent them throughout the design process. They are composite archetypes based on behavioural data gathered from the many actual users encountered in ethnographic interviews.
In addition (and with reference to Copper's definition), I'm not sure why the author of the article assumes personas are characteristics-centred, whereas archetypes are behaviour-centred. A persona will be of little to no use if you take out behavioural elements from it.
My Mac dictionary defines archetype as:
A very typical example of a certain person or thing.
For instance, you can say "The guy we interviewed yesterday is an archetype of an elderly user".
An archetype is someone who exists that fits a set of known characteristics. For a persona, these characteristics are initially unknown and thus derived from research.
Observing vs Interviewing
You can identify behaviour patterns using either. Observations are what centred but they often fail to provide the why. Interviews, on the contrary, are better in revealing the why but may be problematic in revealing the what or how (see Nielsens's famous First Rule of Usability? Don't Listen to Users).
With this in mind, there is some sense saying that observations are more likely to help identifying that person X is a typical example of Y (we typically identify archetypes through observations), while an interview may yield data more appropriate for personas. But this is not set in stone and depends on what exactly is being investigated.