Today on Github, several interesting new repositories have surfaced, covering areas like 3D asset generation, language model self-editing, and AI-generated video detection.

Here's a list of some notable projects:

• Hunyuan3D-2.1: A fully open-source framework for production-ready 3D asset generation with physics-based rendering.
• SEAL: A framework for training language models to self-edit via reinforcement learning.
• CVE-2025-33073: Proof-of-concept exploit for the CVE-2025-33073 NTLM reflection SMB flaw.
• SoundMind: Implementation of SoundMind, using reinforcement learning to enhance logic reasoning in audio-language models.
• SceneCompleter: Achieves 3D-consistent generative novel view synthesis through dense 3D scene completion.
• GenWorld: A large-scale, high-quality dataset and a novel detection model for identifying AI-generated videos.
• valorant-best-cheat: Shows external cheat development for Valorant with features like aimbot, ESP, unlock all focused on stealth and modularity for educational purposes.
• metadspy: Generates reusable DSPy programs from simple YAML specifications.