Search papers, labs, and topics across Lattice.
The paper introduces CharacterFlywheel, an iterative process for refining LLMs in production social chat applications using real-user traffic data and controlled A/B testing. They refined LLaMA 3.1 across 15 generations through data curation, reward modeling, supervised fine-tuning, and reinforcement learning, demonstrating consistent engagement improvements. Results show significant gains in both engagement metrics (up to 8.8% breadth and 19.4% depth) and steerability (instruction following improved to 84.8%).
Real-world social chat deployments reveal that iterative refinement using CharacterFlywheel can boost LLM engagement by nearly 20% and dramatically improve steerability.
This report presents CharacterFlywheel, an iterative flywheel process for improving large language models (LLMs) in production social chat applications across Instagram, WhatsApp, and Messenger. Starting from LLaMA 3.1, we refined models across 15 generations using data from both internal and external real-user traffic. Through continuous deployments from July 2024 to April 2025, we conducted controlled 7-day A/B tests showing consistent engagement improvements: 7 of 8 newly deployed models demonstrated positive lift over the baseline, with the strongest performers achieving up to 8.8% improvement in engagement breadth and 19.4% in engagement depth. We also observed substantial gains in steerability, with instruction following increasing from 59.2% to 84.8% and instruction violations decreasing from 26.6% to 5.8%. We detail the CharacterFlywheel process which integrates data curation, reward modeling to estimate and interpolate the landscape of engagement metrics, supervised fine-tuning (SFT), reinforcement learning (RL), and both offline and online evaluation to ensure reliable progress at each optimization step. We also discuss our methods for overfitting prevention and navigating production dynamics at scale. These contributions advance the scientific rigor and understanding of LLMs in social applications serving millions of users.