You are oversimplifying the issue and solving a "strawman problem. From the abstract of one of the discussed papers [1],
> We give a protocol for producing certifiable randomness from a single _untrusted_ quantum device that is _polynomial-time bounded_. The randomness is certified to be statistically close to uniform from the point of view of any _computationally unbounded quantum adversary_, that may _share entanglement_ with the quantum device.
where I have highlighted the important terms. Your goal is to generate a random string that you trust that your adversary does not have any statistical information about.
* untrusted: the device that you are using to generate the random string is untrusted. You bought it from the market but have not tested it or can't test it with resources on hand.
* polynomial-time bounded: to get a random string of size $n$, you don't want to run the untrusted device for longer than poly($n$)
* computationally unbounded quantum adversary: the adversary is computationally unbounded. If the adversary was more limited computationally, it might be possible to use a simpler protocol to generate your random string.
* share entanglement: this is important. If the adversary is the one who sold you your diode, he might have entangled it in some way with the devices in his own lab, allowing him to get the same measurement results as you did.
Clearly, this is a very difficult problem, but is relevant to when your random number generator comes installed in your laptop from the manufacturer. This problem is also important for understanding the information processing powers of quantum mechanics.
[1] https://arxiv.org/abs/1804.00640 To add some argument from authority, the authors of this Vazirani and Vidick are two of the giants of the field and know what they are talking about.