Editor’s note: This week’s spotlight is on Yazabi – a platform connecting startups with AI talent. Founders and brothers Jeremie and Ed Harris are both physicists seeking to build a bridge between businesses and talented developers in AI. To learn more about Yazabi and get involved, go to yazabi.com or email Ed at [email protected]
Tell us a little bit about yourselves!
Jeremie: We’re brothers from Ottawa, and we’re both physicists. Ed completed his PhD at U of T, and I did a Master’s at U of T and then came back to Ottawa to do a PhD. About halfway through it, I dropped out to work on Yazabi. Also, our friend Kiran Rao is our CTO.
What sparked the idea that turned into what Yazabi is now?
It’s a poorly kept secret that there are no jobs in academia for physicists. Many of them graduate and go off to Wall Street or something finance-related, but that’s not something any physicist actually wants to do. One thing we realized was that physicists are really into data science and machine learning, and that’s how we got into this. I was working at my desk on something related to image recognition, when someone came up to me and asked about it. After explaining a little bit, he asked me if he could help me … for free! Clearly, there was demand for this sort of work. Here you have a busy grad student in physics who’s got courses and TA’ing responsibilities, yet he wanted to get his hands dirty on a project without any pay.
We also knew a ton of startups looking for machine learning talent through our accelerator, the Creative Destruction Lab. There’s this huge struggle to beat Google, Microsoft and other big companies for this kind of talent, but everyone’s focused on poaching people who have PhDs and Masters degrees in machine learning already, rather than going after people who have effective yet informal training who are often even better..
So we decided to do a little engagement experiment on my floor at the Advanced Research Complex at U of O. We just started recruiting people and pretty soon, the whole floor was on the platform. Sometimes you put your finger on something and you realize there’s a pulse there. That was around 15 weeks ago: since then we’ve grown to the point where we’re on 8 different campuses nationwide, and have 50 students on the platform and over a dozen startup projects posted.
In the world of machine learning, what matters isn’t what you know, it’s how fast you can learn.
Because they’re an underutilized pool of talent with high potential. There is this trend now, where companies that focus on machine learning are saying “if you can show me a STEM scientist with self-taught machine learning knowledge, I want to hire that person.” In the world of machine learning, what matters isn’t what you know, it’s how fast you can learn. And people with physics backgrounds have polished mathematical intuition that helps them blow by everyone else when it comes to learning.
For the uninitiated, what is Yazabi and who does it serve?
There are two sides to Yazabi: the startups and the students. It’s a marketplace that connects businesses that don’t have a lot of resources but are hungry for machine learning talent, with people that have the raw potential to work at Google or Apple – but who don’t know it yet. The scientists who discover Yazabi learn how valuable they are, and the startups realize it’s possible to get machine learning talent inexpensively and with zero risk. Students and startups get to know one another through project work, and can go from there.
Yazabi also gives a guarantee: startups don’t have to pay students unless it’s quality work and something good comes out of it.
We want our clients to get big results and be fulfilled – we want them for life, we’re not trying to make a quick buck.
Can you give us an example of a connection you made between 2 parties?
One of our earliest jobs went to a student who was having a hard time in his lab. He was thinking about dropping out of his PhD, but was worried about the risk that represented. He joined the platform, completed our basic skill testing exercises, and we found him a full-time job on a machine learning project within a few weeks.
Now this guy is connected to an awesome company and working remotely and loving it. We want our clients to get big results and be fulfilled – we want them for life, we’re not trying to make a quick buck.
What side-effect of the advancement of machine learning, are people not talking enough about?
People should be talking about the issue of AI ethics. For a lot of people, the idea of self-driving cars for example, is just a curiosity. They don’t really internalize the fact that it’s not just about self-driving cars. Startups are going to replace a large amount of work that doctors and lawyers do. We’re going to automate an awful lot of the jobs that we took as being inaccessible to AI and machine learning.
One of the problems if that happens, is you may find a concentration of wealth, and a disparity of inequality that spreads much more easily. Politically, this should be talked about: how will we deal with AI taking over our jobs?
A lot of laypeople may be thinking about whether or not this can really happen. But they’re out of the loop. This is taken for granted in machine learning circles: it’s going to happen. It’s not a matter of if, but when. The general public should be more aware of that reality.
This plays into how we designed what we’re doing. Machine learning talent often gets scooped up by big companies, which leaves no space for startups who can’t compete. Yazabi is the solution in that respect: a lot of companies talk about democratizing AI, but we’re actually doing it.
Machine learning spreads across a bunch of languages and tools, which ones does Yazabi have experience with or recommend, and why?
We have plenty of experience with Python, TensorFlow, Pandas and scikit-learn. For quick and dirty projects, scikit-learn is great, but when you need more horsepower, TensorFlow is the way to go.
Looking back, where did Yazabi waste the most time in early stages that was unnecessary?
Ed: In the first year, 99% of what we did was a waste of time. For example, we developed technology that we didn’t end up using. It was interesting tech, but in hindsight it wasn’t very useful.
We also wasted a lot of time hiding our idea. We went to ridiculous lengths to do this: not telling our friends what we were doing, not discussing it with our early mentors and advisors, even avoiding the use of email for more “sensitive” exchanges. In hindsight, it’s hard to think of a more pointless investment of energy than that.
Because we didn’t talk about it to people, we didn’t get user feedback. You can’t secretly build something in your cave, and then come out with something and expect everyone to sign up. If they don’t care, they don’t care. And you can’t make someone care; they either do, or they don’t. Your job is to figure out what they care about.
If I could give myself a piece advice 2 years ago, I’d have said: drop out of grad school – you’re wasting your time.
Jeremie: If I could give myself a piece advice 2 years ago, I’d have said: drop out of grad school – you’re wasting your time. It’s seductive: you get a little bit of money so you don’t have to worry about the real world, but it’s a deadly waste of productive years.
To prospective startup founders, I’d say: learn the skills you need for yourself. Don’t pay other people to do the work that founders should be doing. If you don’t know how to build an app, learn how to do it. It took us a while to come to that full realization, but at a certain point we decided if you want it done right and at low cost, you have to do it yourself.
Jeremie, what purchase, of $100 or less, has most positively impacted your life in the last 12 months?
Jeremie: Two things, actually.
A bus ticket to Toronto for a meeting for our accelerator program. We got a huge amount of advice, opportunities to speak to VCs and to connect with a lot of our clients. It also helped us get some face to face time with our market and do real-life recruitment experiments. Just being able to go down there physically and talk to real people can be really valuable.
The second, is The Lean Startup. Buy it. Read it.
Ed, what has been the best investment of time or energy you’ve made in the last 12 months?
Ed: Before Kiran was our CTO, I spent some time working with him to go through a machine learning textbook that recently came out. This was partly to force myself to understand all the material in it, and partly as a way to train him, and partly as a way to evaluate what kind of a person he was. He ended up joining us for a co-op term this semester, and then coming on as our CTO. He’s incredibly gifted. The time that I spent with him was probably the best time investment I’ve made in the last year.
Do you have any interesting sources of inspiration that have helped you think differently about your work?
So many of the strategies in machine learning apply to startups. Surprisingly, as a startup, it is sometimes possible to go from week to week and not have a long-term plan if you just focus on making the right move each week. That’s not an obvious thing. Most people are trained to think for the long-term, and look ahead a lot.
In machine learning, a neural net does the same thing. It takes batches of examples, and improves itself a little bit after each batch. Give it more examples, and it learns more, so on and so forth. You can almost look at AI and machine learning as inspiration for startups. We have this picture in our head of our startup as a neural net that’s slowly climbing up a hill of optimization. It’s a weird source of inspiration, but it’s legitimate.
Talk to real users: do not build a single thing until you have spoken to 10 people.
Any advice for new entrepreneurs?
Don’t always look to hire other people for tech stuff: consider learning it yourself. Often times you understand the vision better than anyone else you can communicate it to, it’s okay to bet on yourself.
Talk to real users: do not build a single thing until you have spoken to 10 people.
We wish something like Yazabi existed when we were younger. One of the things about entrepreneurship is you just have to get off your chair and go do things to learn. Machine learning is one of those skills. You can make money for your startup while learning skills you may need as an entrepreneur.
Also, read Paul Graham – here’s the essay that we recommend starting with: paulgraham.com/start.html