Over 700 AI researchers from around the globe registered for Radical Ventures’ AI Founders Master Class, featuring conversations with AI luminaries about what it takes to make the leap from researcher to entrepreneur. This week we share a segment of our conversation with Aidan Gomez, CEO and Co-Founder of Cohere and one of the creators of Transformers, the foundational model architecture powering generative AI and large language models. Aidan shared his insights on leading a rapidly scaling company over the last three years and delivering impact from machine and deep learning. He spoke with Radical Ventures Managing Partner and Co-Founder Jordan Jacobs. The following excerpt from their discussion is edited for length and clarity.
Jordan Jacobs (JJ): You made the transition to founding a company three years ago. You started your career as a researcher – you were at Google Brain, first in the Valley, then in Toronto. Why would you leave a great position as a researcher?
Aidan Gomez (AG): It was a tough decision. I was walking away from a huge commitment (my PhD) and a great setup at Google Brain where I had amazing collaborators and all of the compute that I could want. The decision came down to asking myself, “How do I maximize impact in the world? How do I give myself the ability to change things in the way that I want them to change?”
I had been at Google Brain for three years at that point and I was part of the team that created the Transformer. I had watched it contribute to translation, then get adopted into search, and increasingly be adopted into other product areas. There was this nagging feeling that there was a bigger project. I wanted to put the tech that we were developing as researchers into the hands of people and see it proliferate in the products that I use as a consumer. After months of discussing with my Co-Founders, eventually, it felt so compelling that it paled in comparison to alternatives.
JJ: Many of the participants in the AI Founders cohort are in the same place you were when you decided to found Cohere. What should they be thinking about if they are considering starting a company?
AG: As a student, there are a few options. You could go into Big Tech to work at one of the big AI labs and contribute to the mission there. Another option is to pursue academia. Or, you can step out on your own and start to build something from scratch.
A lot of people may strongly disagree, but in my opinion, there was a period in machine and deep learning where the highest impact direction you could go was academia and research. I think that period is over. Instead, it’s time to flip into a build mode and think about how we can take the past decade of work since ImageNet in 2012 and actually put applications in front of people and power businesses. I’m seeing the golden era of these big industrial labs doing pure research coming to an end. We are entering a new era of startups that have been enabled by the past decade of progress. I’d love to see a lot more people experimenting in the product space, instead of experimenting purely in the tech space.
There has been a promise that AI is going to change the world and we’ve been waiting for it since 2012. And, in the past two or three years, something has finally changed. We’re seeing technology proliferate into products. Being part of that momentum, starting an industry, that’s the most exciting and high-value thing you can be doing with your time – it’s in the next generation of startups, it’s in building the next generation of products.
JJ: If you could go back to the version of yourself on the day you incorporated what advice would you give that Aidan?
“Operate in a way that assumes success. Assume that you are going to succeed and work backward from there.”
AG: We’ve grown really quickly at Cohere. Often, you’re building the airplane as you’re flying it. There have been some hard moments. I would say, “Trust yourself more.” Operate in a way that assumes success. That is, assume that you are going to succeed and work backward from there. Ask, “What do I need to be doing today to reach that success?” That advice has served me very well.
If you are an AI researcher interested in entrepreneurship, the Radical Ventures AI Founders community offers members an opportunity to join conversations with AI pioneers and access to practical seminars and resources designed to support entrepreneurs looking to commercialize their research.
AI News This Week
-
Machine learning could vastly speed up the search for new metals (MIT Technology Review)
New research suggests that machine learning could help develop metals with useful properties, such as resistance to extreme temperatures and rust. Currently, scientists typically run experiments in the lab to look for ways to combine metals to create new ones. It is a laborious process of trial and error that yields more failures than successes. AI can be used to predict which metal combinations will yield the best results, according to a paper published in Science earlier this month.
-
AlphaFold’s new rival? Meta AI predicts shape of 600 million proteins (Nature)
Earlier this year, almost every protein structure from all known organisms in DNA databases was revealed in DeepMind’s trove of some 220 million predicted protein structures. Scientists at Meta are now using AI to predict the structures of approximately 600 million proteins from bacteria, viruses, and other microorganisms. The model’s speed and accuracy are impressive, but some researchers wonder whether it really outperforms AlphaFold’s precision when it comes to predicting proteins from metagenomic databases.
-
Researchers turn to AI to model how snow cover is shrinking (CBC)
Researchers used AI to compile a long-range picture of how snow cover around the world has changed since the 1980s. Their study, published in Nature, found that globally, snow cover has been decreasing over the past 38 years, with four percent less mountain area covered with snow, and an average of 15 more snow-free days per year. Some of the current limitations with satellite imagery such as low resolution and cloud cover errors are being tackled by Radical Ventures’ portfolio companies Pixxel and Muon Space.
-
Industry Report: Global AI Adoption Index (IBM)
According to IBM’s recently released Global AI Adoption Index 2022, 35% of companies now use AI, up from 31% a year ago. Another 42% of companies said they are now exploring the technology. In the survey, IT senior decision-makers from around the world were asked about AI deployment. Respondents range in size from businesses with fewer than 50 employees to those with more than 1,000. The most popular use cases for AI were automating IT operations, IT or software asset management, activity monitoring, and automating customer care experiences. The top issues inhibiting AI adoption were limited AI skills, expertise, and price.
-
Introducing Cohere Sandbox: Open-Source Libraries to Help Developers Experiment with Language AI (Cohere)
Recent advances in large language models (LLMs) have fueled state-of-the-art performance for NLP applications, such as copy generation, tooling for conversational agents, and semantic search. While LLMs are incredibly powerful, not everyone has first-hand access to them. This week, Radical Ventures portfolio company Cohere launched Sandbox, a collection of open-source repositories to empower the developer community to experiment and work with large language models.
Plus, in Radical Ventures portfolio news: Bill Gates visited Covariant to experience the future of AI robotic automation. Bill challenged the Covariant Brain, the company’s AI Robotics platform, to pick chaotically arranged items.The Covariant Brain did not disappoint, directing the robots to pick with accuracy and ease.
Radical Reads is edited by Leah Morris (Senior Director, Velocity Program, Radical Ventures).
