A new AI training method is designed to boost efficiency and mimic human-like reasoning in large language models. Microsoft researchers claim the approach has the potential to be groundbreaking.
Microsoft Corp. today released a hardware-efficient reasoning model, Phi-4-reasoning-vision-15B, that can process multimodal files such as scientific charts. The model is based on two existing ...