Leaked/announced information about Grok planning 500B parameter model release. Signals continued competition in frontier model scale.
Research on hyperparameter optimization and transfer learning for mixture-of-experts models. Addresses efficient scaling of MoE architectures.
Research on model editing techniques to improve reasoning capabilities. Explores interventions beyond standard fine-tuning.
Research paper on autonomous skill development and evolution in agent systems. Advances methodology for agents to expand their own capabilities.
Theoretical framework applying information theory to LLM capacity and scaling laws. Provides new mathematical lens for understanding model limitations.
YC S24 startup enabling real-time model weight swapping without inference interruption. Novel approach to model serving and experimentation at scale.