Proposes TerminalHorizon, an automated engine that synthesizes long-horizon task environments and agentic trajectories that preserve exploration, evidence accumulation, and iterative improvement; uses ~1.5K environments and ~2.8K trajectories to fine-tune Qwen3.5-27B and improves performance on multiple benchmarks.
Post-training pipeline that boosts game-control transfer by fine-tuning on 6,044 curated gameplay trajectories and then continuing training with PPO across 16 synthetic 2D/3D environments, extending the keyboard/mouse action space with relative pointer motion, sustained inputs, and composite actions.
Generates synthetic German personas grounded in population and geographic distributions for model training and synthetic data generation. The dataset contains 1 million records with six German-language persona types, based on census and other official statistics.