Microsoft Trains Massive Internal Base Model

Microsoft's in house research team has built a base model containing three trillion parameters that matches leading open coding benchmarks while utilizing significantly less compute during training. Conversational and chat tuned versions of the system have not yet been released publicly.