Algorithmic information theory bets that the world is simple: it gives more weight to worlds that shorter programs produce, and physics seems to agree, since its laws fit on a page [1]. I read that as saying a good model of the world should be small. So hearing that a world model rich enough to simulate reality needs an enormous description sounded like a contradiction.
It isn't, because of what "simple" measures. The Kolmogorov complexity of some data is the length of the shortest program that outputs it, and nothing else: not the memory the program uses, not how long it runs. The digits of are the standard example. A few lines of code print all of them, so they are about as simple as data gets, yet getting the trillionth digit still takes a long computation.
The world is simple in the same way. A short program can hold the laws and a simple early universe, but to output today it must simulate 13.8 billion years. Starting from today instead doesn't help: then the program has to hold today's state, which is far from short. Kolmogorov complexity charges for any state written into the program but not for the time spent computing one, so the shortest description of the world trades state for time, and time is free.
Leonid Levin's time-bounded complexity puts the clock back, charging a program for the logarithm of its running time as well as its length [1]. By that measure the world is not simple to predict, and there is no contradiction left: the laws can be short while any model that forecasts with them is large or slow.
References
An Introduction to Kolmogorov Complexity and Its Applications [link]
Li, M. and Vitányi, P., 2019. Springer International Publishing. DOI: 10.1007/978-3-030-11298-1