Topic
All digests tagged Reinforcement Learning
This session covers Reinforcement Learning, detailing the Markov Decision Process (MDP) framework and the Policy Gradient algorithm. Key concepts include defining states, actions, transition dynamics, and using the Bellman equation for optimal value estimation.
Read article
Scaling AI to long horizons requires more than bigger context windows; it demands better environments, robust simulations, and advanced algorithmic techniques like using value models for variance reduction.
Read article
The base model's job is shifting from mirroring web text to building a prior for reasoning and agentic behavior. This requires integrating synthetic, SFT-flavored data early into pre-training.
Read article
Gemini Robotics 2 introduces whole-body intelligence, dexterity, and multi-robot collaboration. The ER model enables robots to understand complex tasks like cleaning a garage using natural language prompts.
Read article
The Kimi K3 model utilizes a complex training pipeline involving multi-tier on-policy distillation across nine specialized domain experts. Key innovations include partial rollout schemes and dynamic resource scheduling.
Read article
GRPO trains agents by sampling multiple completions per prompt and using the group's relative scores as the training signal. This method moves beyond imitation learning, allowing models to learn from their own varied trajectories.
Read article