The meeting breaks. Maddy gathers her things, gives me a look I can't quite read - excitement mixed with something else. She follows Elad and Amit out the door.
Jian and I sit in silence for a moment.
"Seven to fifteen people," he says finally. "By yesterday."
"We should start making calls," I say.
We open our laptops. The tagline stares at me from my email signature, suddenly feeling heavier than it did this morning:
Inferthermic: Intelligence to buffer the entropy Empirically proven 98% Risky Interaction Reduction (RIR)
I start drafting.
Email 1:
Subject: Joining the Alignment Revolution
Hi Sarah,
I saw your work on the Anthropic safety team - particularly your paper on gradient-based monitoring. We're building something at Inferthermic that I think would benefit from your expertise.
We're not trying to solve alignment from first principles. We're solving it from the outside in - post-processing, entropy reduction, making existing models safer without retraining them. It's working. We've got traction, revenue, and a real path to acquisition.
I know you're probably happy at Anthropic. But if you've ever wanted to move faster, break things (safely), and actually ship solutions instead of writing papers about them, let's talk.
Coffee? My treat.
Alex Vukovic CEO, Inferthermic
Inferthermic: Intelligence to buffer the entropy Empirically proven 98% Risky Interaction Reduction (RIR)
Email 2:
Subject: Infrastructure Challenge
Marcus,