Microsoft's Phi-3 shows the surprising power of small, locally run AI language models - Ars Technica

Microsoft’s 3.8B parameter Phi-3 may rival GPT-3.5, signaling a new era of “small language models."