AMA: I'm Phil, I cofounded Element AI with Yoshua Bengio and raised 100M$ in 6 mo, ask me anything

@rob_gold This is an excellent question, and what we've had to ask ourselves on day 1. We believe the dynamics you describe are one of the major reason why we haven't seen the rise of a social network offering a marketplace for third-party algorithms. We believe one of the key step in ushering a more open environment of recommender systems is to demonstrate alternative algorithms can be built. Right now, all the platforms are relying on variation on the same principles: (a) observe user's behavior, (b) infer recommender's parameters to maximize metric-of-the-day, where the metric is seen as a driver of revenue. Waverly breaks this by letting users control their algorithm using natural language. To show the value of such a recommender system, we built a platform for those who suffer the most from the lack of transparency and control on their algorithm: knowledge workers. Our platform seek to empower communities of professionals so they can provide a contextualized stream of content to their followers... and ultimately to the world! We believe it's possible for that platform to be fun and rewarding without being unhealthily addictive. With the right tools, people can integrate good habits into their live — we believe Waverly can be such a tool.

@sneha_saigal In the early days, EAI was a talent play. The biggest needle mover when we started was building a network of very enthusiastic and engaged AI academics who loved us and talked positively about us to their students. Then we managed to hire a lot of very talented AI researchers and AI engineers. Keep in mind that this moved the needle for the raise and for getting into significant consulting mandates with some of the top logos in the world. It did not translate into PMF. If you're curious about this entire story, I talk about it in this PH post.

@philipsnyder I knew I loved to build new things, but Google seemed to offer less and less opportunities for engineers to innovate. At some point I looked at myself in the mirror and thought: this place is comfortable, but it doesn't allow me to be who I truly am. That was the first step towards quitting. The other steps were really, really hard... Ultimately I took the decision more because it felt right more than anything. I can't say it was very rational beyond that... The risk ultimate risk was that, by staying at Google, I felt I might end up regretting how I had decided to live my life. Re: The future of "pratical" AI. Products built on advances in NLU (including generative models), Vision, and speech understanding. We're saying the tip of the iceberg with Dall-E, MidJourney and others. I think these will ultimately usher an era of radically new UXs. We'll gain new powerful ways to (a) express our intentions and (b) express our preferences that were impossible to imagine with previous rigid UIs. There's going to be a lot of exploration in that space. (Waverly is one such experiment.) Eventually, we'll gain new ways of accomplishing many of the tasks we do on computers today.

@kiracheung Most important lesson: Relentlessly focus on a product. I decided to leave when the itch to build Waverly became too intense. I knew this thing could be built with the current NLU tech and I knew it could have a huge impact. I couldn't sit idly and wait for others to build it.

@mingxi Great questions! 1) The hardest lesson was that it was really comfortable at Google, and that some days, as you're out in the jungle of startupland, you'll long for that comfort. Finding ways to remind yourself why you did the move — by sharing it with friends and cofounders — was the best way for me to work through these more difficult patches. 2) Best decision: leaving Google to do something else. Even though it probably wasn't a financially sound decision, it put me on a path where I'm ultimately much happier than I was. Worst decision: Raising too much money pre-PMF. I talk about it in length in this PH thread. 3) With Waverly we're doing product-first, small-team, calculated raises. It's less splashy-in-your-face than "the fastest $100M series-A in Canada" but, ultimately, I believe it has a higher chance of being successful.

@joshfreeland My Element AI experience can't help much here, as we did not really do such things. With Waverly it was mostly about leveraging my networks by sharing our inspiring vision, then converting that into an ever growing mailing list that I kept engaged through monthly email that included some "gif demos". It worked pretty well.

@krzysztof_majewski1 A friend recently told me: « the biggest impact will be the alumnis building great things ». I think it’s right. Sometimes I’m concerned that we will learn the wrong lesson from it and become too shy of trying big risky things… Re: raising money for Waverly. I’ve had the EAI question come up. The challenges were already well known in the investment community when I was raising. It was a good thing that my intention was radically different, with a product-first approach… In my experience investors do value previous experience, even when the result is not ideal. As long as you fail honestly and have a critical look at the decisions you made and which ones you would change.