Skyfall AI

Most AI startups are building harnesses on top of LLMs. The entire AI industry is celebrating how the ‘AI software engineer’ is the breakthrough of our generation. There used to be a time when Yahoo and AOL were the undisputed kings of the internet by being the best curators and organizers of existing information on the web. However, they lost relevance very quickly because they were optimizing for the easy win (curating information manually via humans instead of indexing and searching all the websites). Google’s page rank algorithm quickly became the dominant search tool and paved the path for Google to be the ultimate winner in the race. 30 years have now passed and we are at a similar junction of time again. Instead of focusing on the low hanging fruits of building ‘AI co-pilot’ experience, we want to build completely autonomous businesses end-to-end. We are building the foundation layer of super intelligent enterprises that require little to no human intervention, which we call ‘AI CEO’. The benefits of this are quite obvious and profound - increased productivity for the entire human race so that all of us can focus on our own passions in sports, arts, creativity or spending time with our loved ones. I shared more about Skyfall AI’s next steps in the The New York Times article that came out this weekend. Thank you Sarah Kessler and Andrew Ross Sorkin for the feature.

Most AI startups are building harnesses on top of LLMs. The entire AI industry is celebrating how the ‘AI software engineer’ is the breakthrough of our generation. There used to be a time when Yahoo and AOL were the undisputed kings of the internet by being the best curators and organizers of existing information on the web. However, they lost relevance very quickly because they were optimizing for the easy win (curating information manually via humans instead of indexing and searching all the websites). Google’s page rank algorithm quickly became the dominant search tool and paved the path for Google to be the ultimate winner in the race. 30 years have now passed and we are at a similar junction of time again. Instead of focusing on the low hanging fruits of building ‘AI co-pilot’ experience, we want to build completely autonomous businesses end-to-end. We are building the foundation layer of super intelligent enterprises that require little to no human intervention, which we call ‘AI CEO’. The benefits of this are quite obvious and profound - increased productivity for the entire human race so that all of us can focus on our own passions in sports, arts, creativity or spending time with our loved ones. I shared more about Skyfall AI’s next steps in the The New York Times article that came out this weekend. Thank you Sarah Kessler and Andrew Ross Sorkin for the feature.

Last week, ~40 of us gathered at Stanford for the first Autonomous Organization Summit, hosted by Stanford OpenLab, Stanford AO, and DAOstar. Here's a quick wrap of what happened: 1. We didn't talk much about definitions. An AO is (1) autonomous and (2) an organization. We spent the morning looking at examples from Andon Labs's cafes and radio stations; to Adam B.'s AI Village where ChatGPT Sol and Claude Fable organizes parties together; to Dotta B's 0-person companies; to case studies by Helena Rong from Raft's "Slack-for-agents", to an open source "symbient" managing a French co-living house by Apolinário Passos (Poli) and friends. 2. Axel Wennström: we need to think about "trust injection" as much as prompt injection. People can build a relationship over time with an agent, and then get it to do things. In general, we RLed agents to be assistants, but assistants may not be great at running a business. Just try asking your AI boss for a raise (you'll probably get it). 3. Smarter models will help. But Dotta B's line stuck with me: if you don't trust an agent with your email, that's an infrastructure problem, not a model problem. 4. Measurement really matters. We need better benchmarks, better evaluation suites, and more data in general. Andon is working on harnesses, while Skyfall's Sam Pasupalak presented Morpheus, a big-world benchmark for continual learning in AOs. 5. On the qualitative side: Joel Leibo for DeepMind reframed personhood as a social technology: a bundle of rights and responsibilities we assign for a purpose, not a property we discover. 6. One early consensus: AOs have to act in the real world. For a business, that means tracking P&L. But AOs don't make money... yet. One exception pointed out by Sam McCarthy and others: DAOs on the blockchain, some of them nearly fully automated and running at a billion ARR for years. 7. In a day of cool stuff, Alex Obadia reveal of ARIA's Scaling Trust Arena was perhaps the coolest. The Arena is/will be a real physical place in the UK where multiple AOs operate and offer services directly to people. Not a simulation. I'm definitely launching my AO there. 8. In the last session, we wrote down offers and asks across the field. Roughly, in order: legitimacy, people, money, applications, tools. This is the beginning of a field. Next stop is the London AO Summit this fall. Stay tuned for more information! If you want in: join us at stanfordao.org. Let's start building. I'll post links to recordings in the comments.

It was amazing to share the stage with other AI leaders at the Stanford Autonomous Organization Summit last week and introduce Morpheus, the Big World Benchmark for Reinforcement Learning. Just as human intelligence thrives through continuous adaptation, enterprise AI systems need to learn and evolve in real time. At Skyfall AI, we're building toward Continual Learning from first principles by rethinking the fundamentals. Through Morpheus, we have shown that building a Big World environment is the key to making continual learning a reality, and I'm grateful to have shared our vision at Stanford University. Thank you to everyone who attended and engaged with thoughtful questions. If you haven't tried Morpheus yet, I'd encourage you to explore it: https://lnkd.in/g4VqBnQt Thanks again for the invitation Joshua Tan and our researcher Pavan Seshadri who is the first author of the paper.