By Aaron Rosenberg
This week, Aaron Rosenberg joined as a Partner at Radical Ventures, based in London. Prior to Radical, Aaron was the Head of Strategy & Operations at DeepMind, where he began working directly with the founder, Demis Hassabis, in the Office of the CEO. Previously, Aaron was a management consultant at McKinsey & Company, where he also served as Deputy Chief of Staff to the Global Managing Director Dominic Barton (who is also now a Radical Partner based in London). You can read more about Aaron joining Radical in the Financial Times.
Artificial intelligence is the most transformative technology that humanity will ever invent.
This was my hypothesis when I joined DeepMind in 2017 to work with Demis Hassabis, the co-founder and CEO. Witnessing the rapid research advances we achieved over the years that followed – from game playing (AlphaGo, AlphaZero, AlphaStar, and MuZero), to protein folding prediction (AlphaFold), and conversation (Chinchilla and Sparrow), and more – strengthened that hypothesis into a conviction.
What’s more, a consistent pattern emerged: in each instance, after stretches of intense effort, we saw a sudden takeoff in performance that exceeded our expectations and demonstrated capabilities beyond our imagination. This repetition of positive surprise, reflective of the exponential nature of progress in this field, left me awestruck and optimistic about the trajectory of this technology.
As head of strategy and operations, I had the privilege to contribute to the development of DeepMind as an organization: assembling the executive team, scaling from 500 to 1,500 employees, opening new international offices, establishing ethics governance mechanisms, and leading business planning and resource allocation. Yet the most rewarding part was having a hand in applying our machine learning systems to the world’s most pressing problems, including incubating and spinning out Isomorphic Labs, built upon AlphaFold to redefine drug discovery. Through these experiences, I grew ever more confident that the opportunity to commercialize AI research is overwhelming, far surpassing the capacity of any one organization (or even a handful, however large).
I joined Radical Ventures to realize that opportunity. Radical has been dedicated to supporting AI founders and innovation for years. Having cultivated the Toronto and wider Canadian AI community and extended its presence to the Bay Area, with investments across the US and the globe, Radical has backed an exceptional set of portfolio companies both financially and, critically, operationally through its distinctive Velocity Program. Radical will take London and the surrounding ecosystem, brimming with AI talent, to the next level.
I look forward to expanding the firm’s reach and relevance in this region and to meeting anyone similarly inspired about how AI can and will shape our future. Please get in touch!
Radical Reads for the week of April 2, 2023
1) The Potentially Large Effects of Artificial Intelligence on Economic Growth (Goldman Sachs)
Generative AI has the potential to provide substantial labour-cost savings, raise productivity, and create new jobs, possibly leading to a productivity boom and raising economic growth by 7%. In the US, generative AI adoption could lead to a 1.5 percentage point increase, growth similar to the booming period of the 1950s and ’60s. However, the size of the boost will depend on AI’s future capabilities and adoption timelines. Although rapid deployment of AI could outpace the creation of new roles, the majority of the jobs in the current labour market were created in the past 80 years. The landscape will likely see both winners and losers, but firms investing in AI are poised for better performance.
2) Search Has Its Goliath. Could Richard Socher Be Its David? (The Information)
Richard Socher, CEO and co-founder of You.com, is not your typical tech CEO. When he’s not working on reinventing search, you might find him soaring through the air on a paramotor, a backpack-propeller parachute-like rig. In an interview, Socher drew parallels between his paramotor preparation and the limitations of traditional search engines like Google. Even something as essential as wind conditions has limitations in Google’s one-size-fits-all approach. But with You.com’s open-platform search engine, users have the power to choose relevant apps and tailor their search experience. Socher hopes that You.com can put more control in the hands of users, just like he experienced with his paramotor. Radical led You.com’s Series A.
A letter co-signed by Elon Musk and thousands of others – including Yoshua Bengio and Apple co-founder Steve Wozniak – called for a six-month pause on the development of systems “more powerful” than that of GPT-4. Critics have accused the Future of Life Institute (FLI), the organization behind the letter which is primarily funded by the Musk Foundation, of prioritizing imagined apocalyptic scenarios over more immediate concerns such as bias in AI systems, in some cases so the authors’ own competing startups can have time to catch-up.
4) Explainers: Model Arbitrage (Matt Rickard)
Large language models are especially good at generating new examples. Model arbitrage involves using generative models to generate training data for other models. This can reduce the need for manual annotation and enable models to be trained on large amounts of synthetic data. This approach is not entirely reliable as generated data is not guaranteed to be realistic and diverse.
5) Research: RoboPianist: A Benchmark for High-Dimensional Robot Control (UC Berkeley)
RoboPianist is a new benchmarking system for high-dimensional robotic control that tests precision, coordination, and planning, all with an underactuated system. Designed to master the piano through bi-manual dexterity using a pair of simulated anthropomorphic robot hands, RoboPianist covers a broad range of 150 variable-difficulty songs. RoboPianist offers opportunities for further research in multi-task learning, zero-shot generalization, and multimodal (sound, vision, touch) learning. Pieter Abbeel, one of the benchmark’s creators, is also a co-founder of Covariant, a Radical Ventures portfolio company.