Soldato
I agree with this.
A general developer can learn a language, however there are aspects of development such as security that require a secondary thread of knowledge. Recently I was working with quantum information scientists - they research the maths that feed into algorithms for quantum cryptography. Security IT such as CISO and down need confirmation of knowledge (certifications) for risk management, but in the end they are broad cybersecurity unlike the very specific R&D in that area.
Digital transformation (lipstick on the pig) doesn't require a plethora of degrees, however applying development to scientific computing, chemicals or material research modelling etc, does require an understanding of the field of application.
Having written target operating models and build devops teams, I would say that there is a mindset and skills associated with devops but (sorry to say) it's not rocket science. Like machine learning, the tools and languages anyone can pick up but the fundamental statistical maths and subject matter requires training (ie doing FFTs and not considering spectral leakage due to ignorance). Too many people claim to know tools and libraries - but they don't know the foggiest what they're doing with them.
It would actually be quite interesting to see a breakdown of some of these fields who have/don't have a degree - for example security I know in my experience specifically for red teams, there's generally been very few degrees required as it suits the people who aren't necessarily interested in education, but gained their interest in other ways. Things like algorithms are another interesting one, I see a lot of people taking a huge interest in algorithms outside of Universities nowadays, much more so than 10+ years ago, because of the nature of everyone and their dog copying the FAANG interview process these days of whiteboarding algorithmic and data structure problems, despite probably some crazy statistical number never needing to write a custom bubble sort in their career, but there will always be edge cases. Scientific Computing is an interesting one too - I wonder how many go with a traditional Comp Sci or CSE degree rather than say a Biochemistry degree. I do agree they're probably more likely to require degrees for those fields.
You mention ML and doing things like FFTs, and yeah You'll learn FFTs in a Comp Sci degree, but you probably have to question out of 100 people who did a degree who has ever used a FFT in their career, and again something that should be able to be picked up on the job, unless you get a Grad who has just come out of Uni or something who is interested in the pure maths side of things, let's be honest they're going to have to look FFTs and other advanced Maths up all over again. You're 100% correct though, too many people do claim to know tool and libraries and products without the foggiest of information, interviewed a Principle VMware Engineer a while back who couldn't tell me how vMotion worked, so I see that in my side of the world too. He was degree educated too, but that's a product you don't generally see on degrees so irrelevant.