As far as I understand it, I think chaos theory basically says “there are some functions that depend so strongly on their initial conditions, that a very small difference in the initial conditions can lead to a very different outcome.”
For instance, take the trajectory of a ball on a rectangle billiard table. If you shift the initial conditions (the initial position and speed of the ball) by a very small margin, then you should expect the resulting trajectory to also be changed by a small margin. As a very concrete example, if you shift the initial position of the ball by 1 cm in any direction, and don’t change its initial speed, then the new trajectory will be entirely parallel to the old trajectory, and shifted by exactly 1 cm. So, we can say that there is no chaos on a rectangle billiard table with a single ball, and that trajectory of the ball is easy to predict.
However, if you do the same thing on a billiard table with a round obstacle in the centre, then even a very small shift in the initial conditions can result in a widely different trajectory after the ball has hit the round obstacle. So, with a round billiard table or a round obstacle, there is chaos.

Billiard with round obstacle, from https://www.chaos-math.org/en/chaos-v-billiards.html
This doesn’t mean that the trajectory of the ball cannot be predicted when there is a round obstacle. It just means that the math is more complex, and that the trajectory is a lot more sensitive to the initial conditions. Geometrically, given perfect information about the initial conditions, a middle-school student should have the tools to solve the problem. However, if you account for measurement errors in the initial conditions, like physicists do, then the problem becomes much more complex and the small incertitude on the initial measurements translate into an incertitude so huge in the final position of the ball that the prediction is useless.
Similarly, when you roll a die, you can pretty much assume that the result is fully random. In a sense, it is not, and if you had all the data describing the initial conditions, then it should be possible to predict which face the die would land on. However, the math is very complex, and the result is so sensible to the initial conditions that you’d need extremely precise data; not just the initial position and speed of the die, but also its initial 3d rotation on itself, its elasticity, etc.
Likewise, weather forecasts become very inaccurate if we try to predict the weather more than a few days in the future, because the weather is so sensible to the initial conditions, and the initial conditions include so much data: pressure, temperature, wind speed, etc, at every point of the globe all around the world.
Quantum physics does introduce some unpredictability, for instance in the form of Bell’s theorem or Heisenberg’s uncertainty principle, but I think the impact of this on the observable world have been greatly exaggerated. How hard it is to predict the behaviour of one individual particle is typically not very related to how hard it is to predict the behaviour of a body at a larger scale.