Ep 44: Co-Founder of Together.AI Percy Liang on What’s Next in Research, Reaction to o1 and How AI will Change Simulation

The evolution of AI research centers on the transition from static language models to reasoning-capable agents that utilize test-time compute to solve complex, multi-step problems. Percy Liang, a Stanford professor and co-founder of TogetherAI, argues that academic research must remain orthogonal to frontier labs by prioritizing transparency, auditing, and public-good benchmarks like SciBench. Beyond task-based performance, generative agents offer a new frontier for digital twins, enabling researchers to simulate social dynamics and policy interventions with unprecedented detail. While current models show promise in math and coding, their integration into larger systems requires a shift toward holistic safety and defense rather than mere model-level constraints. Ultimately, the field is moving toward a paradigm where AI acts as a sophisticated co-pilot for scientific discovery and creative expression, necessitating modular architectures that evolve beyond traditional transformer designs.