Breaking

Amazon AGI Scientist Explores Model Fundamentals Behind Both Creative and Effective AI Performance

October 21, 2026 at 12:28 EDT

Foundation Models
Research & Papers

Peter Gostev interviews Pradeep Natarajan, Senior Principal Scientist at Amazon AGI, on a fundamental question: can frontier models excel at both creative and technical tasks, and what makes it possible?

Arena Conversations • October 2026

Mastering Both Creativity and Correctness

Pradeep Natarajan on how post-training fundamentals enable frontier models to excel at technical reasoning while preserving creative capability.

Reasoning Progress

27%→12%

Dissatisfaction rate improvement in math and reasoning tasks

Creative Domains

Plateaued

Writing, gaming, legal reasoning show flat or minimal gains

The Fundamental Challenge

Models optimized for originality drift from accuracy. Models optimized for correctness become formulaic. Post-training recipes must balance competing objectives: curated instruction data, reinforcement learning signals that reward both creativity and correctness, and evaluation methodologies that measure success across all domains.

20+

Years ML expertise

Nova model tiers

Rubric

Dynamic eval criteria

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.