Replit CEO Amjad Masad: general models should JIT-train their own smaller replacements
amasad · x · 2026-10-04
In a clip shared by a16z, Replit CEO Amjad Masad argues the hot topic shouldn't just be recursive self-improvement, but "models training their replacements."
- He draws an analogy to JIT compilers: just as an interpreter spots optimization opportunities while executing dynamic code and emits optimized machine code on the fly, large general models (he cites Operator and Astra) could detect that a use case is narrow and train a small domain-specific model in real time to take over.
- The pitch is "just-in-time" distillation: general models act as discoverers and trainers, crystallizing repetitive narrow workloads into cheaper, faster specialist models.
- He closes with "Who's building this?", framing it as an open startup/research opportunity.
More from AGI Musings
- Debate: Running an LLM Fits in a Warehouse, Duplicating a Brain Would Need a Star — JoshPurtell · 2026-10-04
- Jon Finger: P(doom) measures the proclaimer's imagination, not real AI risk — mrjonfinger · 2026-10-04
- Ethan Mollick turns Sutton's Bitter Lesson into a country-western music video — Afinetheorem · 2026-10-04
- Anthropic's Amanda Askell: three modes of how AI should relate to users — AmandaAskell · 2026-10-04
- AI video is AI's second-biggest PMF as ReelShort heads to $1B+ revenue in 2026 — StewartalsopIII · 2026-10-04
- Sergey Karayev: Your Cells Don't Know You — And You Don't Know What You Comprise — sergeykarayev · 2026-10-04