The Next Episode: I'm joining Magic Pony Technology
If you look at the permalink, you'll notice this is not the first time I wrote a blog post called The Next Episode. Last time was back in March 2014 when I joined the early stage tech investor Balderton Capital as their data scientist. My job was to introduce data-driven thinking into the investment process of a VC firm. I had a chance to work as a member of the investment team, seeing some of the VC processes from the inside, going to partner meetings. I've learnt an awful lot there. But first and foremost I met and worked with so many super smart entrepreneurs, that I had to entirely recalibrate my bar for calling someone smart.
Next gig: Magic Pony Technology
I'm leaving my job in VC-land and joining London-based product-driven machine learning startup Magic Pony Technology as a research scientist. Why? When I started thinking about my next gig should be, I had the following desiderata:
- has to be product-driven as opposed to a blue-sky research funded by investor money, I wanted to build something that solves real problems for real people, right now or at least in the near future. I also wanted the company to be long term sustainable as a business, not something that is acquired quickly by a tech giant. Many people agree we are in a tech bubble, lots of money is available at seed stage and particularly for AI. I think a period of disillusionment is bound to follow. Not being addicted to external funding is extremely important to us. Having a commercially interesting product aligned with the core research is what differentiates SpaceX from the Apollo programme. Many people try to build the Apollo programme for AI, I want to build the SpaceX of AI.
- machine learning has to be a real enabler: I realised that my passion is machine learning, and I'm one of the lucky ones to have come to the field before the buzz around deep learning started, so I think I'm in a pretty good position to find the really smart people and build truly novel things. But I knew I have to work on something where machine learning is crucial, where it enables a true tech revolution, where the barrier to producing results is mainly technological. I have applied machine learning to social data before, turns out, your success mainly depends on whether you have sales momentum, machine learning is a nice to have. Similarly, you think about AI in healthcare? The key barriers are not technological, the tech is pretty much there. The barriers are political and regulatory. With Magic Pony I think we found an area of products that is genuinely enabled by machine learning, and the reason the solution does not exist yet is technological. Stay tuned.
- team: work with only the best people, don't compromise: One thing I learnt working in VC is that the team, especially the founders are a super important factor determining the success of a company. Magic Pony assembled a small, very technical team of pretty impressive people, 9 out of 10 of us have PhDs covering topics from machine learning, image processing, computer vision, computational neuroscience and HPC. It's a cliche to say we don't hire people who are not smarter than us, but I will actually try hard to enforce this policy as it pays off in the future. I already feel dumb when I compare myself to the people who are following me in the team, like Lucas Theis. And again, stay tuned :)
bonus: working on stuff that interests me: I tried to make this the weakest criterion, but after having been away from machine learning research for years, I really felt like I want to work on these problems again. I really wanted to do research again, but goal-driven research. I had a list of what I think the interesting frontiers of machine learning currently are:
- unsupervised learning
- learning from video, exploiting temporal structure
- natural language understanding
- machine learning for control and decision making
My work is now about unsupervised learning for video and visual data, combining two important frontiers, so I couldn't be happier. Particularly as generative modelling, so dear to my heart as a Bayesian, is making a comeback these days.
As I was looking for things to do, in fact looking for a company to start, I realised I already found a place in Magic Pony that satisfies all these criteria. This is the company I want to build, and it's definitely a topic I want to work on. If you want to hear more, get in touch I can always be persuaded to chat over a coffee.