Researchers from UC Berkeley, ICSI, and LBNL have proposed a new approach, LLM2LLM, to enhance Large Language Model (LLM) performance in low-data regimes using synthetic data. Concurrently, the AI community is witnessing significant advancements in LLMs, with the introduction of models like InternLM2 and a new model from MosaicML. InternLM2, an open-source LLM, has shown to outperform its predecessors across six dimensions and 30 benchmarks, designed with a 200k context. Meanwhile, MosaicML has released a new open weight LLM that surpasses Grok-1, LLama2 70B, and Mixtral in general purposes and rivals the best open models in coding. This model is an Mixture of Experts (MoE) with 132B total parameters and 32B active, trained for 12T tokens over 3 months on 3k H100s. These developments indicate a rapid evolution in the field of LLMs, potentially paving the way for Artificial General Intelligence (AGI).