One interesting aspect to this is that most companies focused on AI are chasing the same research talent, which is indeed rare. But sometimes AI research is not the most important skillset, especially when applying something like deep learning to the problems that a business faces. There are many software engineers capable of understanding the math behind deep learning, and piecing together the parts for ETL, training and inference.

Full disclosure: I was quoted in this article.

An average high schooler is able to understand the math involved in DL, but not even the leading researchers can explain why certain things work better than others, let alone predict what will work better.

Well said.

In my experience, model interpretation is an undervalued skill. In most environments you need models that are both accurate and interpretable.

Can someone please explain to me, AI talent that is sought after is... is the search for mathematicians who can _create_ ML tools, or merely use ML tools (like scikit-learn or whatever-have-you, to optimize ad-targeting or things like that)?

I take it most of the times it's the latter. In which the case the AI talent in question doesn't need to be the most hardcore programmer, just someone who knows statistics, can do some basic linear regression and whatnot and knows one of the more popular ML libraries, like scikit-learn or tensorflow.

In my experience, neither. Most are searching for people with feature engineering expertise (i.e. people who know what and how to extract learnable features for a particular problem domain).

"Typical A.I. specialists, including both Ph.D.s fresh out of school and people with less education and just a few years of experience, can be paid from $300,000 to $500,000 a year or more in salary and company stock, according to nine people who work for major tech companies or have entertained job offers from them."

Wow, half a million dollars in salary, I'm living under a rock.

Sign me up!

It's less than you think in the Bay Area. Above average engineers easily break 300k. Really good ones will break 500K.

After rent and taxes in the Bay area though, if you're in the 300k bucket, you'll only have 140k left not including higher food / transportation / private school costs. Definitely a good amount but much less impressive than the original sounds.

Only 140k left over!? How do they eat?

Free food ;-)

Don't forget taxes paid on illiquid stock compensation.

Mileage definitely varies on living expenses. Anyone who doesn't have a family (yet) can save far more than that.

Sign up to move to Silicon Valley. I don't think they're paying these salaries in other areas.

Still, articles like these do make me regret leaving AI and moving into web development 15 years ago. Back then, there wasn't a lot of work in AI.

I work at one of the companies mentioned in the article although not doing ML. From what I can tell, these stories do a disservice since all these people come into the company wanting to do ML when most work is not nearly so glamorous.

Grab it while you can, it won't last, and then there will be a load of people on market with inflated salary expectations.

I have learned to use Tensorflow Object Detection API, learned how to augment images and am yet to find an object I cannot detect.

Will this be true of other fields like chatbots? Where all I need to understand is how to train and inference?

I work on image segmentation and object detection research. In fact, my work can be found in the classifiers provided by the TensorFlow Object Detection package. I’ll be blunt, we suck at object detection and we really suck at object detection where there’re more than a handful of classes (there’s a reason we use top-k for evaluation) which is most problems. Furthermore, we’re entirely useless when it comes to problems that requires some interpretable taxonomy (e.g. make and model of a car). I encourage you to try some harder problems. :)

Really? I have an image with 180 classes on, so far (slight model config changes from pets tutorial) I've managed to detect 80% of the classes. Don't get me wrong images are similar in both sequence of objects and background (supermarket shelf). I have a budget to bring in a consultant, is this something you'd be interested in?

Is there any way for me to start getting up-to-speed and possibly being hirable as an AI/Deep Learning expert (with alot of luck) in the next year or 2, without going back for a MS or PHD? Where should I start looking, in terms of guidance and discussion? I don't have a family yet, so I have lots of free time outside of work still.

There's a nice list here: https://news.ycombinator.com/item?id=12900448

There have been some great resources released since that list was written. For the courses, the exercises/assignments are the most important part.

- Deeplearning.ai's courses on Coursera (course 4 of 5 coming out next week)

- Practical Deep Learning for Coders Parts 1 & 2

- Book: Hands-on machine learning with TF and SKLearn (Aurelion Geron)

If you want a really gentle path, I'd start with 'Data Science from Scratch' by Joel Grus.

If you want to start at the deep end, buy the Deep Learning Book by Goodfellow et al. They review the relevant math at the beginning, but it's work to go through it. Perhaps if you recently got out of school you'll find it easier than others.

What's the salary for a new graduate (Bachelor's) in this field? Do they even get the opportunity or are these jobs only available for Masters/PH.D.s?

These are research positions, so it's unlikely undergrads are getting in on that.

As a cs undergrad, i found it hard to get more than an intro level skillset in AI. Machine learning, nlp, algorithms for data science were all offered as grad courses, which i was able to take, but the undergrad offerings were underwhelmingly basic