The opportunity for early stage AI companies
- The age of the research-led AI startup is coming to an end unless you are one of a handful of teams / leaders that can credibly present an approach (and possess a differentiated ability to attract talent and capital) that positions you as the third general AI pole alongside and . Commercialization needs to be a core part of the early thesis for almost every company.
- We underrate the value of marginal gains deep inside of enterprise and industrial workflows where big tech and general AI companies won't compete (ex. 1% reduction in error rate). Small gains can cascade/compound both in terms of unlocking new business growth opportunities and in systemically impacting technology capability. The 20% of "deep value" that comes from 80% of the work/time (building on top of generalized systems) should be the focus of most startups.
The cultural differences between OpenAI and DeepMind
How significant has the role of geography and culture been in the relative successes of the two companies?
- OpenAI has taken big risks betting on a few key projects. They have leaned into the scaling challenge by finding something that is working and scaling that one thing as fast as possible (GPT). The "blitzscaling" approach has helped them essentially catch up with DeepMind despite starting later and not having Google backing and the narrative that emerges has a reflexive impact on the org overall.
- DeepMind has been extremely effective at attracting talent through an almost hybrid academia approach (akin to early semiconductor companies) — fusing the best of a broad range of academic disciplines with "entrepreneurial energy". This will continue to attract high level talent and will be a key driver in forcing academic institutions to re-invent themselves in order to compete.
More broadly, despite the accelerating success of these two companies, we have been slow as societies to truly grapple with the governance related challenges posed by these superintelligence projects.