My previous thesis was that you should just simplify things and people will watch. That is, something is theoretically really complex and you simplify it, and that’s your innovation, and people appreciate it. And definitely they do, but the scale is just out of whack.
See, intuition works, but you need to fine-tune other things. Otherwise, you will spend so much cognitive resources on simplifying really tough things and then realize that, hey, there is no demand for this product I just made.
For example, there exist things like ReLU, GELU, and their need. Okay? You can explain that intuitively—you need to transform numbers to properly compute them for LLMs at each stage, right? You really explain that. But how much is even the demand?
Because to get to that point, you are way too deep in the sauce. It’s like, first you have to understand LLMs a bit in general, then you have to understand vectors passing through the transformer layers, and when they’re being transformed, what are the operations being performed?
And this is where our ReLU and GELU come in, that hey, I am explaining why we transform them. It’s like way too deep. It’s way too bottom of the funnel type of thing.
Basically, the topic you choose to simplify determines the engagement you receive. This is why idioms or sayings are so popular that they generalize at such a higher level that they can appeal to everyone.
And if you are trying to appeal to everyone, you will have to do a lot of context building. And even for those familiar, you still need some context building. So the deeper you are in the source, the more context building is required.
And context building is bad for us because it takes a lot of cognitive effort.
Plus, it needs to be independent of other things. It can be enjoyed without having contact to anything else. It should be a standalone type of thing.
And within itself, it should be standalone that you go to any place and you don’t feel out of whack. That the context chaining is not so stringent and necessary that if you skip 10 seconds, you just don’t understand the video anymore.
This independence from both other things and itself allows people to enjoy things and interact more with them. This is exactly the reason why you will see people spending so much time on Instagram Reels than a book. On Instagram Reels, open any reel and you can enjoy it independent of other reels.
And even if you skip like 10-15 seconds, everything is so simple that you can just catch on to it. But on the other hand, if you look at a technical book, no one wants to touch them these days because the context chaining is so necessary and stringent that even if you zone out for a few pages, the next part of the book is just inaccessible.
You better have a proper amount of intuition. You better have the minimum amount of context needed to understand it. It better have high independence from itself and other content.
And it better have good payoff or promise of payoff because most engagement comes from first 5 seconds.
So the TLDR is: have minimum context rate, have maximum independence, both from your own video and other videos, and try to be intuitive.
This is for reels btw, maybe context debt and dependence is why people might wanna watch long form video and even for short form this filters out alot of topics you can talk about