Launching Airtree's Frontier Program
A residency for idea-stage founders: Community, Capital, Compute
Excited to officially launch Frontier - a residency for idea-stage founders.
We think the fastest way for an exceptional founder to find the right thing to work on is to put them in a room with five or six peers of the same calibre who'll raise the ceiling on what they think is reasonable to attempt.
Alongside the community, we’re giving founders time to navigate the idea maze with $250k on an uncapped MFN SAFE, and we’ve partnered with the major labs, infra and cloud providers to accelerate teams with credits.
More info and the link to apply here
On my mind
Enterprises are beginning to complain about AI spend at the same time as post-training infrastructure becomes more widely accessible (Tinker, Baseten Loops, Prime Intellect). In the next 12 months, I expect larger companies to begin fine-tuning their own models for high-volume applications that don’t require frontier capability.
But designing good reward functions for domain-specific tasks requires genuine expertise - signals are ambiguous, gameable, and degrade across long-horizon tasks. RL environments are also complex to construct and maintain. The tooling to do this without an ML research team doesn’t really exist yet.
But these are solvable problems. And in a world where enterprises consume orders of magnitude more tokens than they do today, bringing down the unit cost of that intelligence is important.
Keen to hear from others thinking about this, particularly anyone who’s tried to build reward functions outside of coding or maths.
Links
For those interested in the bull and bear cases for biotech as AI accelerates:
How fast could robot production scale up?
What’s changed in financial markets in the last decade?
This encapsulates the nihilist sentiment I feel every time I go to SF:
“If you were to glance out one day and see a row of mushroom clouds rising on the horizon, you would know at once that what you were seeing, remarkable as it was, was intrinsically not worth remarking. No use running to tell anyone. Significant as it was, it did not matter a whit. For what is significance? It is significance for people. No people, no significance. This is all I have to tell you.”
Tracks
Raaj shared this in the Airtree newsletter this week, gorgeous: