Jerry Tworek on Leaving OpenAI: Research, Risk, and AI's Future

Departure from OpenAI

• 💡 Jerry Tworek, a key figure in reasoning-driven models, recently left OpenAI after seven years, seeking greater research freedom.
• 🎯 His decision was influenced by OpenAI's appetite for risk and org chart limitations that hindered cross-organizational research.
• 🧠 Tworek felt his vision for the path forward in AI research diverged from the company's chosen direction.

State of AI Research & Innovation

• ⚠️ Tworek expresses sadness over the homogeneity among major AI labs, which are largely pursuing the same transformer-scaling approach.
• 🚀 He believes the industry needs to move beyond current transformer architectures and pre-training scaling for true breakthroughs.
• 🔑 He championed reinforcement learning (RL) on top of language models at OpenAI, demonstrating new capabilities beyond pre-training alone.

Future Research Directions

• ✨ Tworek is most excited about two underinvested areas: architectural innovation beyond transformers and continual learning.
• 🧠 Continual learning, where models learn continuously like humans, is seen as a crucial step before achieving AGI.
• 📈 He updated his AGI timelines to potentially 2026-2029, contingent on advancements in continual learning and multimodal perception.

Reflections on the AI Industry

• 💬 He acknowledges the fear and responsibility felt by researchers when developing significantly new AI technologies, such as the