You can form a hypothesis without basing it on anything. You could for example randomly generate 1 billion sentences and then try to test if they are true.
This is not what is meant by "hypothesis" in Occam's razor, which is about hypotheses that are based on actual assumptions (and using these assumptions to pick a "best" hypothesis).
Okay, it sounds like what you call assumptions, I would call "data". Or "background data" or something.
If I think Bruce Wayne is Superman, I might base that on the fact that they're both physically very fit; that one would need to be very rich in order to have the kind of technology that is indistinguishable from alien powers; that Bruce Wayne's parents were murdered, and this could conceivably draw him to a life of fighting crime, which is a thing Superman does.
That sort of thing leads me to form the hypothesis: "Bruce Wayne is Superman".
But that sort of thing isn't what Occam's razor is about. It's about things that we haven't observed to be true, but which would need to be true for the hypothesis to hold. You should prefer a hypothesis that requires fewer such things.
If I see Bruce Wayne and Superman in the same room, then in order for Bruce Wayne to be Superman, he must have an identical twin. I haven't observed him to have one, but that's what the hypothesis requires. Accordingly, my confidence in the hypothesis decreases.
The initial hypothesis is only a starting point. When building a model where 'mice are smarter than humans' you need to account for all the evidence out there.* Compared to the model where 'humans are smarter than mice' it's vastly more complex or vastly less testable.
* I have heard this defined as hypothetical baggage or implicit baggage. ie. if CO2 is not increasing temperature then why not?
