Andrew Ng's VC fund will use their own teams instead of listening to pitches. What can we read from this? (1) the market is saturated with 'wannabe startups' that talk about ML but don't have a clue (2) VCs get many of these pitches 'X but with AI' with noone knowing ML in the team.
Andrew has the reputation to do this and not be hated. But you know who did exactly this? Rocket internet, the clone-war masters.
"In his view, AI businesses are also different from regular startups because you generally get a closed feedback loop that allows you to quickly see what works (and what doesn’t)."
What is he talking about here? First of all, this is just the basic lean startup methodology. Any startup (especially consumer ones) try to do that, i.e. A/B testing new features. Second, he may be referring to the case that once you have a big comprehensive dataset already gathered, then you can just iterate improving the model.
My (cynical) answer: good luck with that. The real issue with AI especially in mission-critical situations is data in edge cases. Even if you can improve on your train/test data, that barely translates to customer satisfaction unless the edge cases are treated well. And edge cases will be revealed very slowly by encountering data in the field. Just ask anybody working on voice-assistant agents (Alexa, Siri, Google one) or self-driving. It's not a fast loop by any means.
Of course, Ng probably knows all this which makes me feel even more bothered.
>>Of course, Ng probably knows all this which makes me feel even more bothered.
Ng's implied claim is that he has a head start on how to tackle the edge cases, along with other hard problems encountered in the field, based on his experience at Baidu. This is what got him the "easy" capital raise. If he really has an edge (no pun intended), then it makes sense for him to apply it while it's still relevant.
Maybe. Part of me thinks it's somewhat of a small money grab. (hope it's not.)
What I will bet money on though is that no massive success will come out of this. The key to AI is data & hardware (and not algorithm or methodology -- where everyone is essentially doing the same) and that's the domain of already giant tech companies. There can be a lot of good use cases in teaching far-off industries on how to use AI (manufacturing, etc.) but that's essentially a consulting business which will have a limited window. So lots of little successes is completely achievable; massive successes I hardly doubt.
I am surprised it took so long for this to happen. Ideally speaking you would want the famous ML guy to help you in vetting any startup claiming to solve the problem using AI/ML.
“One of my philosophies of building companies is the importance of velocity,”
True. And the key to keep (increasing) velocity is momentum. I believe the fundamentals to that is infrastructure. Once you get to the "flywheel stage" of an AI company (http://nicodjimenez.github.io/2018/01/25/stages.html), experimentation becomes so easy that new models can be rolled out nightly.
Actually the key to keep increasing velocity is maintaining a positive dv/dt, that is to say, acceleration. Momentum is more of a tendency to refuse to leave a state (think of `inertia-in-motion`), rather than actively seek it.
I think goal-driven is what keeps the momentum, regardless of the amount of investment and machinery available (though I have to admit a clean desk is better than a messy desk).
Is it worth learning deep learning with coursera,?I was considering doing the MIT deep learning path on YouTube. I'm not really focused on AI I just want an overview.
> Final as in this will be the last time Ng will teach the course?
Courses on Coursera aren't really taught that way. The course instructor(s) prepare the course materials (videos, presentations, quizzes, assignments, etc.) once only. Courses start on a regular schedule and the platform, rather than the instructors, takes you through the material in sequence. There are forums with moderators for peer-support, but you don't typically interact with the instructors as they're not really there.
I assume the OP is referring to Ng building the fifth of the five courses that comprise the specialisation.
Maybe there's room for multiple players in this space, but I imagine the name recognition and bona fides of Andrew Ng is going to suck the air out of All Turtles.
I have a feeling that Andrew's fund is much more focused on cutting-edge AI than Phil's. If you take a look at Phil's portfolio, you see can see that the companies are probably using some form of AI but that's not the core of them. Also, in my personal experience with talking to Phil, it seems that for him the "practical AI" part is more important than what kind of tech is used. So, maybe a more consumer-focused perspective (which I like).
I can imagine that Andrew focuses more on companies that have or have the potential to have significant AI themselves. I doubt that Andrew would consider a bot that uses the Google Vision API and Dialogflow an "AI" company. If that makes sense.
PS: I do think that the latter should be considered an AI company. Just like a company using AWS to host is called web company. In fact, probably a lot of companies will be using some sort of AI api pretty soon.
Sounds interesting! Does anyone have examples of AI startups that solve real world problems? Just trying to get a picture of what companies would fit their fund.
I think mapping roads is a real world problem. But for a few urban areas, maps are rarely updated. Of course Google has some sort of monopoly here, but there is a LOT of work to be done.
Planet labs now have the ability to map the entire planet every day at a 3m resolution. So object detection applied on these images can be a quite efficient. I still wouldn't call this AI as we are talking mostly about supervised learning but it's a highly practical real world use case. CrowdAI and the like are on this path already.
I'm not sure if this falls (strictly speaking) under AI but lately I have been using Google Maps for Traffic almost daily. It shows the congested roads, estimates the time, suggests paths and show length, etc...
It is battle tested by me and I can say it is surprisingly accurate.
I mean these are great stuff. But do you plan to be a venture-backed company (i.e. looking for hyper-growth and an exit, e.g. acquisition or IPO) or are you planning on being some sort of LLP?
The problem is that though the above may currently be a profitable business, I can't see how you will generate a "moat" and mature into a monopoly or a big market share in a vertical.
I'm a YC founder with 7 mill in funding and 40 employees. We already are :).
Skymind is a weird mix of joint ventures in asia, US and Chinese investment. We're not looking for an exit anytime soon. We are looking to build a big business though.
The "moat" is a land and expand strategy. There is lockin with our tooling. It's a standard on prem play. We're hard to get rid of once you install us. We help in house teams compete with external vendors. Interest actually aligns there. Happy to elaborate a bit otherwise.
No we're a horizontal AI platform vendor with a strong focus on anomaly detection in time series applications. Our main product is a competitor to AWS sagemaker for on prem and hybrid cloud deployments.
My startup, RAMM Science (https://ramm.science) has a Deep Learning Platform, we have customers using Deep Learning for:
Predicting real estate opportunities (which houses are about to sell) predicting energy usage for thousands of sites in real-time, predicting churn, predicting Ad-Tech prices and fraud in real-time, predicting anomalies in seismic radar scans, customer segmentation and recommender engines.
Those are examples of real world paying customers using Deep Learning
Manufacturing is a big one -- right now most factories are still run very manually, but advances in robotics, computer vision, and AI will change the game. In terms of companies here in the bay area working on this, theres Andrew's own Landing.ai as well as Instrumental (https://www.forbes.com/sites/aarontilley/2017/06/22/instrume...).
As other fellow commenters remarked, and as an AI enthusiast/researcher myself, I fail to see AI as a business.
What I would like is a fund, or similar that is funding research, because fundamentally we need so much research in this area.
AI as we have it now is just a bunch of ML algos and *NNs that perform well in different niches and, well, after all cat pictures were categorized, what to do next?
So, in case Andrew is reading this post, I would like to have a chat with him about my research ideas and if they could be funded. Because if this happens, many more would follow and then we can see a real progres towards real/hard AI.
There are tons and tons of meaningful uses for the current state of ML. Just about every industry is full of opportunities to apply the most basic algorithms and drive tremendous value. It goes without saying that research is still valuable, but what I haven't seen yet is large, widespread use of applied ML.
Of course ML has his uses, even now I am doing fully industrial ML. But, in the end, is just an algorithm/NN that does what your suppose to programm it. And is a heck of manual work involved.
AI is on a different scale andmay have less to do with the actual implementations.
The amount of money in this field puzzles me. The difference between relative novice and competition winner on kaggle.com is often in the single digit percentages or less. When 26 year old George Hotz can build his own self driving car in his garage, it seems like a very narrow moat.
But the point is, for $3M I can fund George Hotz and have an entry in the self driving car arena. For nothing, I can go grab open source code and be a small percentage away from the absolute state of the art.
How do companies justify such a tiny edge as being worth so much? I wish Andrew Ng the best. I do. He's a class act. I enjoyed his course years ago, but came away a little disappointed. At the end, I realized there's no intelligence in the algorithms. It's all just applied statistics and a lot of sweat preparing training data.
Maybe I should just shut up and vie for some of the easy money sloshing around. I think I would just feel weird asking for a ton of cash when the end product is just a bit of recursive math.
I don't think anything other than ML will be funded like that for a while.
We are on the very top of that hype curve, a few of years from now we'll forget how absolutely stupefied we were when ML gave us models that "can X better than humans" and the enthusiasm will give way to a feeling of "I overpaid for this". It happened before with databases way back when, then with specialist systems, then ...
The only difference is now the layperson hears about it with an astounding frequency. I don't know, maybe that will make things different but I can't see it having any other effect than making that 'disappointment crash' harder
I'm really enjoying his ML class on Coursera. Considering taking the deep learning series next. (And I work more than full time at a tech giant. It's working into my schedule okay so far.)
I'm at the end of a PhD. Once I've this baby wrapped up I'm jumping on that course quicker than you can say "rise of the robots". I've heard only good things about it :)
What is the domain of your PhD? If it's computer science or mathematics then you might find that ML class on Coursera too easy, way below your level of competence.
The deep learning series is great. It's a nice start. You'll just have to do a fair bit of extra legwork on your own if you want to be solid on the theory.
Just finished the first four courses, didn’t realize the last course on RNNs isn’t live yet! Hoping he finishes it up soon, the other courses were superb.
Andrew has the reputation to do this and not be hated. But you know who did exactly this? Rocket internet, the clone-war masters.