SotA (State of the Art)

In artificial intelligence and machine learning, "SOTA" is used to describe the best-performing method or model on a specific benchmark or task at a given time. Researchers constantly develop new architectures and algorithms, aiming to improve performance in areas like computer vision, natural language processing, or reinforcement learning. Achieving SOTA typically means that a model outperforms all previous models on widely accepted benchmarks like ImageNet for image classification or GLUE for natural language understanding. SOTA models often represent cutting-edge advancements and can signify key shifts in capabilities, such as increased accuracy, efficiency, or generalizability across tasks.
The term "state of the art" has been in use since the early 20th century, but in AI, it became prominent in the 1990s and gained significant traction in the 2010s as machine learning competitions and benchmarks standardized performance comparisons.
No single individual coined the term "SOTA," as it comes from general engineering practice. However, influential groups like Google AI, OpenAI, DeepMind, and individual researchers such as Geoffrey Hinton and Yann LeCun have been pivotal in producing SOTA models across AI subfields, including deep learning and neural networks.