Creating Agents that Co-Create — Karina Nguyen, OpenAI

Creating Agents that Co-Create — Karina Nguyen, OpenAI

---

---

AI Summary

Summary of Video: Scaling Paradigms in AI Research

1. Speaker Introduction
   • Karina, AI researcher at OpenAI, previously at Antarik.
2. Overview of Scaling Paradigms
   • Discussion of two main scaling paradigms in AI over the past few years:
     • Next Token Prediction (Pre-training)
       • Model learns world understanding by predicting the next word.
       • Automates learning various tasks, e.g., translation, geography.
       • Difficulty in creative writing and complex problem-solving tasks.
     • Post-training
       • Enhancements made post-pre-training to improve function understanding.
       • Introduction of GitHub Copilot as a notable application.
3. Reinforcement Learning & Chain of Thought
   • New approaches in reinforcement learning allow models to think through complex problems.
   • Importance of interactive paradigms that facilitate real-time collaboration with users.
4. Future of AI
   • Transition towards co-innovation where AI works as a collaborator.
   • Exploration of novel interfaces enabling users to create tools without coding experience.
   • Importance of personalized learning experiences.and collaborative data generation.
5. Concluding Thoughts
   • Excitement for AI’s future in becoming a component of everyday software creation.
   • Potential for creating a multimodal, personalized approach to information access.