The thing is, I need someone who understands why the theoretical approaches fail. Someone who can look at our empirical results and help me understand why the math works, not just that it works. Someone who can bridge the gap between the safety research community and the commercial systems actually being deployed.
Your background in formal verification and your skepticism about current benchmarks - that's exactly what we need. The academic world will still be there in five years. The window to shape how commercial AI systems handle safety is closing now.
One conversation. I'll show you our data. You can tell me why I'm wrong.
Alex Vukovic CEO, Inferthermic
Inferthermic: Intelligence to buffer the entropy Empirically proven 98% Risky Interaction Reduction (RIR)
Email 14 (from Jian):
Subject: Leaving the Tower
Professor Volkov,
We met briefly at the Simons Institute workshop - you asked pointed questions about our latency guarantees during the Q&A. I appreciated that. Most academics treat industry presentations as marketing rather than engaging with the technical substance.
You're right to be skeptical. Most "AI safety" companies are selling snake oil. Keyword filters wrapped in venture capital buzzwords.
We're different. I can show you the math. I can show you the production logs. I can show you the adversarial evaluation results that no one else in industry is willing to publish because they're terrified of what they reveal about current LLM vulnerabilities.