Good sources of knowledge I am sure. For me, learning something like C++ is something I need to balance with actually developing product code. For instance the value of something like auto wouldn't be readily apparent without seeing how it usefully applied in product code. Admittedly ymmv.

I wasn't making a claim that it is better, per se. I was pointing out the fallacy of person 2's thinking.

It's hard to explain all of this without getting into deeply personal stories I don't care to share, but I've been self-teaching since I was a kid due to being far removed from classical education / upbringing. I've also worked in many disparate industries, managed to be team leader in many of them, and never once have I had meaningful training.

Does that mean that my way is the best way? Not at all. It simply meant that I found myself in situations where either a) the base knowledge wasn't there at all, and management decided I can figure it out, or b) people with knowledge deemed me capable enough regardless of my actual background.

If your fear is that someone like me lacks humility, I'd argue that we are probably more humiliated with our learning path than average, which is why we study for years before doing it.

I self-learn out of necessity. I've talked to a few people who are person 2, and not a single one did it because they want to. Not a single one is proud of the circumstances that made it impossible for them to go through a classical route. They each decided it was better to use that mental capacity to at least try something than, idk, buying a TV for video games or doing meth. The fact is, each one has a very complicated background, and none of them do it for selfish ego-stroking reasons.

Learning resources - I've always believed in finding books, classes, etc, with homework and, most important, answers.

I'm a firm believer in mastering the fundamentals. That gives you an intuition in what is happening later on, and it gives you the intuition to judge a resource, or at least think on it. Even people with many years of experience and college have a flimsy grasp of the fundamentals.

There is an upper bound to book-learning, and there is a knowledge / practicum ceiling, and that is okay.

The second proxy: does it work the way it should work (this step is not before the fundamentals).

This is a valid concern.

I don't think anyone is claiming that a self-taught is going to build up a new OS or the next Google.

A self-taught can figure out to build a web app, use a database, or admin a server without knowing how to build RoR, MySQL, or Linux by hand.

Bull****. No idea in the history of ideas has ever been stolen and executed on.

There is a large amount of downright wrong chosen answers on SE. There is a ton horrible code on github.

Sure; they are hardly alone and you can access lots of stuff on arxiv, conference papers, personal blog posts, papers, company blogs, whatever; I was simply responding to jjshabado's comments on credentialing that universities are not where that knowledge is made these days (because they lack, e.g. the data and infrastructure).

But I guess the discussion has drifted a bit from what makes sense to me. If you asked me: "I want to become a top ML practitionner. Should I watch a bunch of youtube videos and study on my own or accept Deepmind's job offer?"

I don't think it's even close. In fact, and this is my main point, I don't think it's close even for "reasonably technical startup that has some interesting and business critical machine learning problems to solve".

It's more than this though. If you think you know X, but are totally missing concepts A, B, and C - you probably don't know to ask about A, B, and C.

"Real big data jobs" probably isn't the best example. But even here I know some Universities offer a courses where you get to use an actual cluster of machines and write jobs that actually need to run on say more than 3 computers.

Other people run something like Hadoop on their laptop, run word count on it for a few GB of data, and think they've learned something useful about "Big Data".

The problem with working for Deepmind is that the employee will have to sign an NDA and won't be able to use the secret knowledge in side projects anyway.

Not at all. My only point was that this (the part I've quoted) part of your story doesn't sound exceptional at all to me; it is, indeed, close to the normal course of things in my experience. Someone has to be the first person to learn that stuff at a given company; some people like doing that, others don't.

(What's unusual is the lack of formal education *before* entering the workforce.)

I think that, overall, we are making the same points. I agree that joining a stealth op with some interesting ML is far superior than watching a bunch of YouTube videos for the next 3 years. I also think it is much easier to break in early than late, which is the crux of my entire posting on this topic.

My only disagreement is the idea that someone who self-learns is some stubborn lone wolf. I think this is a poor conception of the reality of many people's situations. Not everyone goes to good high schools, can get into college, can afford a bootcamp, grew up in stable families, etc. There is a lot of complexity to humanity, and at the end of the day, most people are doing are trying to create the best life they can, often without mentorship or guidance. In contrast, to become good at self-learning takes an incredible amount of self-reflection and criticism, along with years of practice.

I don't think the type that will self-teach before classical education is doing it because they genuinely enjoy it. Rather than labeling them as people who "do it their way, on their own" I'd challenge you to consider that some people have little to no choice, and at least did something instead of nothing. In my personal experience, I haven't seen a correlation between quality or type of education and difficulty to work with, willingness to learn, etc.

Try teaching what you think you have learned to another person and see if you can answer their questions. If you don't have a person to teach, see if you can find and answer a Stack Overflow or similar.

Yup.

Fair enough; I was mostly arguing the "superior than watching a bunch of YouTube videos for the next 3 years" angle there. Credentialed/not; willingness to learn on one's own/not; willingness to "be an employee" not are somewhat independent.

I don't agree with this--at the entry level, I do think it makes a difference--but after a year or two of work, it's typically irrelevant or insignificant which is why I advocate for getting that experience sooner rather than later.

This is a good idea.

I don't think that's correct. There are a ton of incorrect answers on SO and there are a ton of poorly thought-out and downright wrong "how to program" blog posts, books, and resources.

Look at what Codecademy became, for example. It was a good idea if you consider it from the perspective of mastery by teaching, but so many of the exercises are far out of bounds, even with a built-in code checkers. This only creates more poorly informed students.

The trap is thinking that you are leaned enough to correctly teach. There are many many examples outside of programming where the "teachers" b.s.'ed themselves to the point they feel correct propagating their misguided beliefs, based on a flimsy knowledge.

I believe that an important role of any teacher is having the humility to know they aren't quite the correct person. This sort of thinking does an incredible disservice to the teacher, who can end up reinforcing their own beliefs, as well as the student who doesn't know any better.

CodeAcademy does suck. And there are bad Amazon books that capitalize on what's popular and shouldn't even make the presses. BUT, there are still plenty of good books on Amazon.

While that may be true, if you post an incorrect answer in such a forum, there is more chance it will be corrected than if you just read a book and think you know it.

However, I was referring to something a little broader which I, perhaps didn't explain. When you try to explain a newly learned concept to someone (in person or on the internet) it often means you have to think more deeply about it or re-examine your knowledge in some new way. At the risk of sounding all Zen, it's that inner reflection that can help you assess whether you have competency in that topic.

It's not perfect by any means, but neither is, for example, taking an exam in a subject.

Yeah. Teaching something to someone can definitely be an eye opening experience for if you really know something or not.

Try writing it down. You lie to yourself in your head a lot...

Studies have shown time and time again that writing down the wrong information actually reinforced that wrong information. Just because there is deep thought, it is limited by your understanding of the subject.

To give an extreme example, I believe Deepak Chopra truly believes he understands quantum mechanics.

When writing it down you are often forced to spell out details that you skip in your head because they seem obvious, and in that process it's quite common to discover inconsistencies/mistakes. You never typed out a post here only to recognize that your thought process/argument is flawed?

Grading the result was implied...

That's the idea.

Sure, and I've pressed "submit" plenty of times where I was certain I was correct even though I was wrong.

The problem I'm having with the advice of teaching to mastery is that there are, idk, 4 bad resources for every 1 good resource. This is a TON of bad forum / SO answers, blog posts, books, YouTube videos, and paid resources that is doing an incredible amount of damage to the students. These students are sadly, going to teach others.

The evidence does not support the idea that writing it down shows people inconsistencies. Just because you and a select population can, it doesn't mean that it works universally. The actual evidence shows it, and the often hilarious studies demonstrate the exact opposite.

Has anyone here ever done development with Salesforce? Just finished up an interview and I think I'm likely to get an offer. I really don't know anything about it but am willing to learn, just hoping to hear a few positives and negatives if anyone has experience.

So I think I read somewhere a CS degree is only 1 year of courses relevant to CS after you remove math classes (trigonometry, statistics,etc). Is that accurate? If so, can you enroll in just those classes?

I don't think you could do that and get a degree. They may let you attend classes a la carte but you need all of the required curriculum to get the piece of paper which is the only part that matters.

Plus some of those math classes might provide insight into the areas you are interested in developing software.

I've done a very minimal amount and it was irritating. But I didn't take time to really understand it so I might just not have understood the system at a level of detail to make it not that annoying.

This was my experience as well. I seem to recall the documentation being pretty weak too, especially if there is anything custom about your setup.