Martin Casado: labs missed Jev because 'beings speak' — software needed a model that chooses
multiply_matrix · x · 2026-09-27
a16z's Martin Casado (joined by Box CEO Aaron Levie and Steven Sinofsky) explains why frontier labs missed Jev:
- Why they missed it: LLMs are text-in, text-out, born from chat — and "we've spent years cramming a thing that spits out text into a traditional program... it's just been super janky." Jev's insight: generating text is expensive and more than you need — give it a set of options and it picks the best one, incredibly fast and cheap, and more accurate because it's trained just for that. "Probably the fastest adoption of an AI model since ChatGPT."
- The core divide: "The labs are trying to create beings, and beings speak... If you're trying to create God, God speaks in natural language" — but software needed a model that chooses, not one that speaks.
- On regulation: most AI regulation debate happens before anyone has defined the risks. Agents don't tire, run at enormous scale, and probe systems like employees never could — which may force rethinking permissions, authentication, and the security stack itself.
- Closing: AI innovation may increasingly happen outside the frontier labs.
Related event: a16z Podcast with Box CEO: Don't Regulate AI Before Risks Are Defined(5 posts)→
More from AGI Musings
- Not Boring: the three arguments that AI isn't a bubble are exactly how bubbles form — Not Boring (Packy McCormick) · 2026-11-11
- Viral post: five years of 'maybe next year,' then 2026 arrived — hey_abusiddik · 2026-11-11
- Jeremy Howard: LLMs "cosplay understanding," like the man in Searle's Chinese room — Machine Learning Street Talk · 2026-11-11
- OpenAI board member Paul Christiano: an AI accident is your most likely cause of death — notkilleveryoneist · 2026-11-11
- 'Chat with your data' misses the point — people want a smart coworker — vboykis · 2026-11-11
- Ben Todd: AI task time horizons appear superexponential in AECI, not just revenue — ben_j_todd · 2026-11-11