Jev: System One models for Prod, not God — with Diogo Almeida, CEO, TypeSafe AI

TypeSafe bets that calibrated, code-callable "system one" models—not bigger chat/reasoning LLMs—unlock automation by prioritizing reliability and data curation over compute scale.

• Jev launches viral: TypeSafe's launch video hit ~40M views in under a week, surpassing GPT-4o's 22M, and Diogo says tokens-per-day already surpassed a key internal milestone with 24/7 machine-driven (not just human) traffic.
• New training paradigm: Jev uses unpublished 'Reinforcement Learning for Calibrated Decisions' (RLCD), a third North Star after RLHF (please humans, causes sycophancy/mode collapse) and RLVR (optimize benchmarks, causes jaggedness).

Themes, sentiment and reasons are AI-written paraphrases, checked so they never quote an episode. They may contain errors. Not investment advice. Episodes remain the property of their creators.