What do we usually refer to as generative AI?

Generative AI is a subfield of artificial intelligence focused on models that learn the underlying patterns and distributions from a training dataset to create new, synthetic data. Unlike discriminative models that classify or predict based on input data, generative models are designed for content creation. This new content can include text, images, audio, code, and other complex data formats that are original yet resemble the data on which the models were trained. The core function is generation, not analysis, classification, or optimization.

B. This describes Automated Machine Learning (AutoML), which focuses on automating the process of building and selecting models, often for discriminative tasks like classification.

C. This refers to model optimization techniques (e.g., pruning, quantization), which aim to improve the performance and reduce the computational cost of any AI model, not just generative ones.

D. This is the definition of discriminative AI or data analytics, which involves learning boundaries to classify or predict outcomes from existing data, rather than creating new data.

1. NVIDIA Official Documentation: In the "What Is Generative AI?" guide, NVIDIA defines it as: "Generative AI is a type of artificial intelligence that can create new content and ideas, including conversations, stories, images, videos, and music." This directly supports the concept of generating new and original data.

Source: NVIDIA, "What Is Generative AI? A Guide." NVIDIA Blogs, May 23, 2023. Section: "What Is Generative AI?".

2. University Courseware: Stanford University's Human-Centered Artificial Intelligence (HAI) institute explains: "Generative AI refers to deep-learning models that can generate high-quality text, images, and other content based on the data they were trained on." This emphasizes the creation of new content.

Source: Stanford University HAI, "What Is Generative AI?" AI Index Report, 2023. Section: "Generative AI: A New Era".

3. Peer-reviewed Academic Publications: A foundational survey on the topic states that generative models aim to learn the true data distribution to generate new samples. "The goal of a generative model is to learn a probability distribution pmodel(x) that is as similar as possible to the true data distribution pdata(x)."

Source: Goodfellow, I. (2016). NIPS 2016 Tutorial: Generative Adversarial Networks. arXiv:1701.00160. Section 2: "Generative Models". Page 2.