humans&: creating AI-native human relationships

No important problem was ever solved alone. Groups —communities, companies, political movements—have always pushed the frontier of human progress. Our ability to live and create with one another, for all of its messiness and difficulty, is what sets us apart as a species. We’re more than just a social species; we are the social species. From the printing press to airplanes to social media, the tools that most transformed our world are the ones that most dramatically expanded our ability to work and live together.

For AI to reach its full potential, it should strengthen our communities, deepen our connections to one another, and make working together better and easier. Achieving that requires a new approach to AI: new models and new ways for people to interact with them, both together and alone, with co-creation at the center. humans& is a human-centric frontier lab built to deliver on that vision. 

humans& does this by expanding the aperture on reinforcement learning. Current models are caught in short-term learning loops, with existing reward and credit systems training them to pursue the most immediate “right” outcome. But human groups’ efficacy are predicated on long-term performance, with complex, intersecting consequences to their actions. humans& is rebuilding models with long-term reinforcement learning at their core, to create AI that’s cohesive and reliable as a long-term partner, and able to plan and set goals like we do. 

Today, we’re excited to share that Felicis has invested in the humans& seed round. The founding team is one of the densest talent clusters we’ve ever encountered, with leaders from xAI, Anthropic, Google DeepMind, OpenAI, Meta, Reflection, AI2, Stanford, and MIT. They’ve defined excellence in both AI research and product, and we believe it’s that intersection of expertise that will power this new paradigm. 

It’s fitting that a company rooted in human connection feels quite personal to me. I met Georges Harik three decades ago, when he was my engineering manager at Silicon Graphics. He went to Google as employee seven, and I followed him soon after. Georges has always been a singular builder and thinker, but he’s also an incredible friend. When he co-authored Quiet-STaR with Eric Zelikman and several of their co-founders, I could tell they were chasing something genuinely new. There’s no one I’m more excited to partner with again than Georges.

This is the beginning of a new chapter of AI: one built for groups of people, in all of their complexity. And if humans& succeeds, it’ll be a new chapter for all of us as well.