Huh, I don't know if it was just me, but I could see the INFP in the video's claim as being far more limited. He just claimed he knew "something" was going on. And he said the would let it resonate with him so he could honor it. I didn't hear him say he knew exactly what the "something" was, or whether the "something" he picked up was the true, hidden, beating heart of the matter, invisible to lesser mortals. Granted, he wasn't being specific, so maybe in any particular case he would feel that way... just pointing out he didn't say so.
Warning: What follows is a series of geeky analogies. I'm a geek, so I make no apologies. I do apologize for the length, though.
I've been thinking about this Fi pretension/assumption vs. Fe, and it seems like some of it could be to the different dynamic we see in Fi vs Fe. Note that where I say "Fe" and "Fi" below, I'm partially referring to "Ne as filtered through Fi" and "Ni as filtered through Fe." I'm aware that Fi and Fe are judging functions, but clearly they have a huge affect on what we pay attention to. I also am not denying the SFPs use Fi, as well, etc, etc.
Active Fe
To me, it seems like Fe works a lot like radar or sonar. Fe tends out an emotional ping ("You're awesome!") and the target(s) respond accordingly ("Thanks! You are, too!"). The social rituals we associate with Fe work the same way. I send you a Christmas card, you send me one. I say you are a friend, you visit me when I'm sick.
While in some sense this can appear as keeping track of social credit (a charge sometimes leveled against Fe), it's also making sure that pings get acknowledged, and lets people know they are valued and where they stand in relation to others. Knowing whose turn it is makes sure that communication continues. Social expectations help make it more clear who should initiate the next cycle.
I see this on the small scale, immediate scale with my Fe friends. They are much more likely to proactively emote ("I'm so excited about this movie coming out!") and like it when you can share with them ("Really? Me, too!"). They also are more likely to reflect what you say back at you, and react to it directly. If you are excited about something, they like to be excited, too, and part of that is reflecting it back at you so you can be sure they are on your wavelength. You don't have to guess that they are excited, they let you know. That gives you a chance to correct them if they are incorrect.
As a programmer, this seems a lot like network protocols that require an "acknowledgement" when the other end has received the data. Some also have both "ack" (for acknowledged) and "nack" (for explicitly NOT acknowledging, so error handling or data resending can happen). These can be used with things like checksums, to make sure that the data received was the same as the data sent. Fe emotional communication seems to have a lot of "acks" and "nacks" going on.
Passive Fi
So, compared to the actively radar/sonar type emotional detection of Fe, Fi works more like a passive sensor. For example, a thermometer or the light sensors in a digital camera. These kinds of sensors respond to external changes, but they don't have to send out a signal first, nor is an acknowledgement an inherent part of their functioning. Like a digital camera, what they pick up may be distorted in all kinds of way (lens distortion, smudge on the glass, noise from the sensor itself, etc), but there's also no intrusive protocol and communication doesn't impose any requirements on the target's (or other party's) part.
So, if someone had only been exposed to radar/sonar types of active detectors, they would be mystified by the function of passive detectors. "How can you passively just get data? What signal did you send out first? What, no signal sent out? Then how did you know anything was going on? You can't have!"
Also, if you were only familiar with ack/nack style communications, you would be appalled at the idea of placing any credence at all on protocols that lacked them. "What? How could you trust the signal that your old analog TV picked up was what was sent? There was no acknowledgement! No checksum! It's totally unreliable!"
So that's kind of how I feel people are reacting when they talk about "pretentious Fi." It's true that there's not all the active sending out of signals and all the explicit acknowledgements and reflection going on. That doesn't mean it's impossible that any data got picked up.
Fi users still have to figure out how to validate their perceptions, just like every other type. I think it's harder for Fi users, since their typical emotional information gathering style doesn't have all the built-in failsafes.
I think it can also set off other types who claim "You can't know that, because there was no communication! I never pinged or acked, therefore you can't know ANYTHING about my emotional state. If you do claim to, you are claiming magical insight!"
I think there can be added irritation from people used to an Fe style of emotional interaction. They ping an emotion at us... we soak it in, but don't really acknowledge it or reflect it back (or at least not consistently). This can can lead to a reaction of, "What, you think you are too good to respond? But later you claim to know my emotional state? So freaking arrogant!" and also "That wasn't communication because you weren't actively involved! How can I know how we (or I) feel if you don't reflect it back at me?!"
To make matters worse, we may go along with something in such a way that WOULD be a real "ack" for an Fe user. Then the Fe user feels betrayed because we appeared to agree when we really didn't.