LLM Future Is Multi-Model | Erik Davtyan #JSConfAM25
summary
Erik Davtyan argues that instead of one AGI, the LLM ecosystem will fragment into many specialized models and infrastructure will need to route and adapt across them. He presents an open router proxy that selects among models like GPT-5, GPT-5 Nano, and GPT-3.5, an intelligent routing layer that directs trivial inputs to cheaper models, and a prompt-adaptation pipeline using genetic/meta-optimization over ~150 iterations that improves accuracy and reduces inference cost by up to 80%, with metrics and a working demo.
AI-written summary · about
chapters (5)
- 0:00Introduction
- 7:13Multi-model infrastructure need
- 11:54Intelligent routing
- 20:12Prompt adaptation
- 29:47Prompt optimization & Q&A
related talks
36mLLMs and AI agents demystified
20marchive – Ethan Ferdosi
12mKeynote: How LLMs Have Changed the Way Organizations Do AI
27mThe Cutting Edge of AI: From LLMs to Multimodal Intelligence- 12mKeynote: How LLMs Have Changed the Way Organizations Do AI
12mThe Future of AI is Specialized, Why General Models Aren't Enough
30mLLMs: Beyond the Hype - A Practical Journey to Scale
01h 39mMastering LLMs: From Prompt Engineering to Agentic AI