Yes, you are right to point this out. There are some important details that are still being debated. Personally my impression is that the debate has advanced enough to the point where MWI can’t be outright dismissed based on this argument. There are multiple plausible explanations and the remaining difficulties have more to do with philosophy than physics.
Edit: To give one example of an approach that I think is promising: We start by describing the observer and environment through a density matrix (a probability distribution over possible wave functions) and introduce an interaction with a quantum system (e.g. a spin). Given a reasonable interaction, you can show that the entanglement in the combined state (observer, environment and spin) leads to the system approaching a state that is a probability distribution of entangled states where each probability corresponds to the Born rule. Interestingly in this case the probabilities emerge from our lack of knowledge about the microstate of the observer/environment, so it’s actually thermodynamic uncertainty.
In the Everett/Many Worlds interpretation the appearance of randomness can be explained as an emergent phenomenon resulting from not being able to predict which part of the wave function we will end up in before running an experiment.
It seems that the JAX developers are focusing their time on making the core framework better and are leaving the task of building high-level APIs to the community for now.
I suspect we'll see a few high-level APIs emerge over the next few months that explore different approaches before the community settles on a particular one.
I really like JAX as well: https://github.com/google/jax.
It's younger than PyTorch and TF, but feels cleaner and more expressive. It has a very nice autodiff implementation (based on https://github.com/HIPS/autograd) and performance is comparable to TF in my experience.
Do you also write your own automatic differentiation tools? Using libraries like TF and PyTorch makes sense if you use neural networks because they provide automatic differentiation (who wants to write out their gradients by hand?) and standard neural network components.
Edit: If your algorithm is not using neural networks, then libraries like TF may or may not be a good fit, it depends on the algorithm.
Writing custom low-level code can still make sense in those cases.
I don't think that the GP is advocating that we motivate students to study linear algebra by telling them it will make them rich. They were merely saying that this is a nice way to grab student's attention.
I really liked it when we went over PageRank in one my university lectures. It made me think "Wow, I can actually have an impact using the things I've learned here".
That's a good point, but I suspect that a lot of serious gotchas that a data scientist might encounter in the wild are not taught as part of a graduate-level statistics course.
Being able to think critically and quickly adapting to the problem at hand might end up being more important than previous experience in stats (which is still very valuable, of course).
Why should not having a BS in statistics prevent anyone from properly learning and applying statistics in a rigorous scientific fashion?
There are a lot of people with a rigorous mathematical background (mathematicians, physicists, biologists, computer scientists, ...) who are perfectly capable of understanding and applying stats concepts at a high level.
In addition, these people have a lot of experience with doing scientific research, so shouldn't they be even more qualified to call themselves "data scientists"?
Can you give an example of something that clearly distinguishes a "data scientist" from say a physicist who learned regression from a stats textbook?
Great question!
There is more to it than you can see in the animation.
Because this is a quantum mechanical problem, it's not enough to take into account a single given field configuration at a given point in time, but we have to work with all possible field configurations at the same time.
Because the number of configurations grows exponentially with the complexity of the system, we have to invest enormous amounts of memory and computing power into simulating it.
Feynman actually wrote this pretty amazing paper on the topic of simulating quantum mechanical systems with computers: http://link.springer.com/article/10.1007/BF02650179.
You should have a look, it's quite readable!
(If the paywall is a problem for you, you should be able to find the paper elsewhere)
Edit: To give one example of an approach that I think is promising: We start by describing the observer and environment through a density matrix (a probability distribution over possible wave functions) and introduce an interaction with a quantum system (e.g. a spin). Given a reasonable interaction, you can show that the entanglement in the combined state (observer, environment and spin) leads to the system approaching a state that is a probability distribution of entangled states where each probability corresponds to the Born rule. Interestingly in this case the probabilities emerge from our lack of knowledge about the microstate of the observer/environment, so it’s actually thermodynamic uncertainty.