Why a New Class of AI “Judgment Models” Could Have Big Business Implications
Summary
This episode explores TypeSafe’s new “judgment model,” Jev, which outputs probabilities for specific questions rather than generating text, making it 20-200x faster and cheaper than LLMs. The discussion covers how this technology enables rapid, low-cost decision-making in workflows like customer support and lead scoring, acting as a “code linter” for knowledge work. The show also reviews recent AI safety debates, including Mark Zuckerberg’s argument for self-regulation and a bipartisan push for human-centric AI regulations, alongside Salesforce’s new enterprise AI announcements.
Topics discussed
Intro: AI Judgment Models and Show Announcements
Zuckerberg's stance on AI safety and pacing
Public reactions to Zuckerberg's comments
Sanders and Bannon call for human-centric AI regulation
Political discourse and Salesforce Dreamforce safety views
Salesforce launches KoA model and AI Force initiative
Sponsors: KPMG, Blitzzy, Section, and Hyperagent
Typesafe releases Jev: A new AI judgment model
How Jev works: Probabilities vs. text generation
Business use cases: Support, sales, and marketing
Jev as a 'code linter' for knowledge work
Jev as a UX for classical machine learning
Multiplayer AI: Teamwork and cross-team commitments
Conclusion: Jev's role in the future model stack
Listen ad-free on Castria