According to Kye Gomez (@KyeGomezB), OpenMythos is an open-source, first-principles reconstruction of Claude Mythos that implements a looped transformer with Mixture-of-Experts routing to enable iterative depth via weight sharing and conditional expert activation, targeting improved efficiency and multi-step reasoning (as reported on X/Twitter). According to Kye Gomez, a community training run indicated OpenMythos achieved its best validation in 2.67× fewer steps than nanoGPT, suggesting faster convergence in early experiments (as reported on X/Twitter). According to Kye Gomez, the team is pretraining 3B and exploring 5B parameter models on the FineWeb-Edu dataset on Hugging Face, followed by GRPO and high-quality RL fine-tuning, with all artifacts to be open-sourced and training scripts available on GitHub (as reported on X/Twitter). According to Kye Gomez, this is an early-stage research effort and a theoretical hypothesis of how Claude Mythos may function, inviting community contributions to evaluate looped transformer models and MoE routing impacts on reasoning (as reported on X/Twitter).

According to DeepLearning.AI, the instructors of the 'Reinforcement Fine-Tuning LLMs with GRPO' course are hosting a live AMA to discuss practical applications of reinforcement fine-tuning in large language models (LLMs). The session aims to provide real-world insights on how Generalized Reward Policy Optimization (GRPO) can be leveraged to enhance LLM performance, improve response accuracy, and optimize models for specific business objectives. This live AMA presents a valuable opportunity for AI professionals and businesses to learn about advanced methods for customizing AI solutions, ultimately enabling the deployment of more adaptive and efficient AI systems in industries such as finance, healthcare, and customer service (source: DeepLearning.AI Twitter, June 13, 2025).

According to DeepLearning.AI, a new short course on Reinforcement Fine-Tuning LLMs with GRPO introduces practical training methods for large language models to improve complex reasoning abilities. The course focuses on using GRPO (Generalized Reinforcement Policy Optimization) to fine-tune LLMs, enabling them to perform advanced reasoning tasks such as mathematics problem-solving, code generation, and games like Wordle without the need for massive datasets. This development addresses a key challenge in the AI industry—making LLMs more efficient and capable for enterprise and research applications. As cited by DeepLearning.AI, mastering GRPO-based reinforcement training opens new business opportunities for building specialized AI solutions that require logical reasoning and decision-making capabilities. (Source: DeepLearning.AI, Twitter, May 21, 2025)