- Stanford Professor
- Founding lead of the Google Brain project
- Author of one of the most famous and loved MOOC's
- Head of AI for Baidu, built up the AI team in both China and the US.
- Founder of Coursera
And I see from wikipedia that he and I are roughly the same age :(
Engineers are often seen as a cost center for most businesses, which means you'll eventually hit a compensation ceiling If you want to elevate yourself into one of the engineers you hear about that is able to break through the compensation ceiling then the below is one of the best ways to do so.....
> My team birthed one new business unit per year each of the last two years
If you can directly tie yourself to a Pnl then you'll always have more options than someone who is considered a cost center.
I hope that what ever he does, he takes some time of first if he needs it. I'd hate for someone like this to get burnt out.
Some excerpts:
Anti-Evil AI drives funding: https://youtu.be/21EiKfQYZXc?t=37m20s
Radiologist will be impacted: https://youtu.be/21EiKfQYZXc?t=57m38s
Against Basic Income: https://youtu.be/21EiKfQYZXc?t=1h1m57s
Trolley Problem (who dies): https://youtu.be/21EiKfQYZXc?t=1h25m50s
In case you can't tell, there's a great big sarcasm tag there. There is absolutely no way that the guy has contributed substantially to all of those things in less than six years.
Sometimes, people get a reputation for something, and then leverage that into a slingshot of career advancement. Good for him, but the people who are doing the actual work of developing this stuff can easily spend six years on a single problem, so it's helpful to maintain perspective.
So ... these are real skills that supersede individual-contributor skills.
In the same way that software offers leverage, really good people-finding -training and -inspiring skills allow you to turn yourself into a group intelligence.
Also note: Andrew's wife is the founder of Drive.ai (and the other two founders are his students).
Like Nicola Tesla ... it's no coincidence he says AI is the new electricity.
This is an underrated point, and something I don't think most people outside of the high level Machine Learning/AI world take seriously.
It's also one of the biggest challenges for the industry going forward because of natural monopolies. I say that because if AI is electricity then data is the coal/oil that drives it.
The big technology players have a massive advantage in their ability to build and deploy tools that collect the data, and then bring it back to be turned into "electricity." If we aren't careful they will be the only groups who can make progress and show actual real world ML driven capabilities - making the barriers to entry even higher.
If you just look at the computer vision space, to do really good Machine Vision you need a LOT of novel image data and the primary platforms creating image content are largely owned by the top 5 players in the form of collection (smartphones) and warehousing (cloud servers).
I'm not sure if there are solutions here that make it possible for a lot of companies to do really well - everyone will just be bought up or out competed by the bigs once the big ones notice a threat on the horizon.
Data is important now, but when we have solved vision, speech, text and robotics to a decent degree, data won't matter as much. The great thing about AI is that we can cheaply copy already trained models or already labeled datasets. There aren't so many datasets needed to solve the most interesting and financially profitable few problems. Of course, there will always be fringe projects where more data is needed, but the main applications will be in the commons. You can copy an AI model if you can talk to it (use it to produce sample outputs). Any model could be copied in a dataset and transferred into another model. The great thing about machine learning is that it learns directly from data, so it's cheap to copy by tracing the inputs and outputs of other public AIs, just as current AIs are taught by tracing the inputs and outputs of people (supervision).
In the former case we are taking data, labeling it, then using it to build our nets and models. You are correct to an extent that it's a usable model once trained and that data is less important.
However, equally if not more important is the data that is being put into the net to come out as a result/action. Arguably this data comes through the same pipe as training data - and the pipes are similarly limited. So its ALWAYS important because you can't take an action or classify or otherwise without it.
When you add in the reinforcement mechanism, or later unsupervised techniques then those data mechanisms blur between training and action data so the point is moot. It's not a one run process in the long run, it's iterative and always evolving based on the user.
"The AI on the horizon looks more like Amazon Web Services — ... This common utility will serve you as much IQ as you want but no more than you need. You’ll simply plug into the grid and get AI as if it was electricity. It will enliven objects, much as electricity did more than a century past."
It's not just AI. Here's Jeff Bezos talking about the internet as electrification
https://www.ted.com/talks/jeff_bezos_on_the_next_web_innovat...
seems a rather common metaphor for a sweeping wave of technological change.
I can't think of other areas where it'll have such a big impact. Robotics (other than Automated driving) has decades more left to fulfill its promise.
1. Do Practical Deep Learning For Coders:-
http://course.fast.ai/ To take a plunge directly into deep learning AI, this course has rave reviews. This course will allow you to do practical industry level stuff first, then learn theory behind it, rather than other way.
Course description:
>This 7-week course is designed for anyone with at least a year of coding experience, and some memory of high-school math. You will start with step one—learning how to get a GPU server online suitable for deep learning—and go all the way through to creating state of the art, highly practical, models for computer vision, natural language processing, and recommendation systems. There are around 20 hours of lessons, and you should plan to spend around 10 hours a week for 7 weeks to complete the material. The course is based on lessons recorded during the first certificate course at The Data Institute at USF. Part 2 will be taught at the Data Institute from Feb 27, 2017, and will be available online around May 2017.
2. Read http://www.deeplearningbook.org/ to gain the relevant math behind it. If you don't know some of the math like calculus or linear algebra presented in the book, learn as you read it from sources like Khan academy.
Now we are up to date on practical side of things, especially deep learning part. We can move on to gain a more generalized and rigorous outlook on various machine learning techniques.
3. Do https://see.stanford.edu/Course/CS229/ - CS229 By Andrew NG , its more rigorous, and complete compared to coursera course. And coursera course is not
I think within a year (max) just this coursework plan would give a strong foundations on practical, theoretical side of things in AI.
I get distracted trying to learn so many stuff at once, (clojure , sicp, haskell, advanced algo) etc etc. So I made this lesson plan to follow as I am interested in AI the most.
And for the practicals, I'll also suggest going over some of the hands-on examples at blog.algorithmia.com, especially if you have some Web dev experience.
>I will also explore new ways to support all of you in the global AI community, so that we can all work together to bring this AI-powered society to fruition.
It is true that AI is the new electricity which will change practically everything in our lives and it is good to see that alliances like OpenAI are forming to democratize the knowledge, this is because giant companies hold a monopoly, they are the only ones who have the sufficient data to do any meaningful research at all.
Very true! I hope he joins OpenAI and helps achieve their mission [1]. It would be a huge boost to their efforts.
Baidu is now one of the few companies with world-class expertise
in every major AI area: speech, NLP, computer vision, machine learning,
knowledge graph.
There's a lot of room to apply machine learning to solve actual problems that many people have, but often its unclear that doing so would end up with results that are significantly better than traditional approaches; or how to achieve those results, tangibly.
I'm sure we haven't heard the last of Andrew Ng; there are a lot of people who want those sorts of skills.
If Deep Learning can approximate any function, does that mean that in some cases it ends up approximating the output of traditional approaches?
I also saw a FB thread that seemed to suggest this may have been a surprise to some of his colleagues. Take that with a grain of salt though - I don't know the people on the thread and they were vague. I just got the sense it was news to them.
“Andrew joined Baidu because of our shared pursuit for the future of AI,” stated Baidu on its official Weibo account, China’s answer to Twitter. “We still have this goal, which is to push forward the development of AI and make life in the future more beautiful.”
“Despite our regrets, we send our thanks and blessings! We wish Andrew even greater success in the future, and hope all goes well!”
Without question, Baidu's dominance over search plays two very painful roles: 1) they're the cutting edge of China's online civilian spy apparatus, and 2) they're the gatekeeper for news and dissemination of gov't propaganda. If you search the web or visit links sponsored by Baidu, either you will find what the government wants you to find or you will be identified as an outlier, and at best, labeled as a potential threat and tracked. It's impossible to imagine that Baidu does not place the names of web denizens matching certain profiles on threat databases thus altering the trajectory of their lives accordingly (and invisibly).
Personally, I can't imagine anyone of conscience working long for an employer so committed to diminish the freedoms of thought and speech. I applaud Dr Ng for his departure.
Correct title could have been "Closing the current chapter...".
OTOH Ng specifically thanked Lu Qi on twitter.
Oh well, I guess eventually we'll know more.