Summary
Highlights
Generative AI is a type of artificial intelligence that creates various types of content, including text, imagery, audio, and synthetic data. It's a key subset of deep learning, utilizing artificial neural networks to process both labeled and unlabeled data using supervised, unsupervised, and semi-supervised methods.
AI is a discipline of computer science focused on creating intelligent agents that can reason, learn, and act autonomously. Machine learning is a subfield of AI where models learn from input data to make predictions. Deep learning is a type of machine learning that uses artificial neural networks, inspired by the human brain, to process complex patterns and can utilize both labeled and unlabeled data (semi-supervised learning).
Supervised learning uses labeled data to predict future values, like predicting tip amounts based on bill size. Unsupervised learning deals with unlabeled data to discover natural groupings or clusters, such as grouping employees by tenure and income. Supervised models minimize 'error' by adjusting predictions closer to actual values.
Discriminative models classify or predict labels for data points by learning the relationship between features and labels (e.g., classifying an image as a 'dog' or 'cat'). Generative models, however, generate new data instances based on a learned probability distribution (e.g., generating a new image of a dog). Generative AI output is often natural language, audio, or images, whereas non-Generative AI typically outputs numbers, classes, or probabilities.
The Generative AI process takes training code, labeled data, and unlabeled data across all types to build 'foundation models'. These models can then generate new content like text, code, images, audio, and video. This signifies a shift from traditional programming and neural networks to models that can create novel content based on learned patterns.
Generative AI learns from existing content to create new content, building a statistical model. When prompted, it uses this model to predict expected responses and generate new content. Large Language Models are pattern-matching systems that predict what comes next in a sequence. The power of Generative AI comes from transformers, which revolutionized Natural Language Processing, though they can sometimes produce 'hallucinations' (nonsensical or incorrect output).
Prompt design is crucial for generating desired output from Large Language Models (LLMs). Various generative AI model types exist based on input and output: Text-to-Text (e.g., language translation), Text-to-Image (e.g., creating images from descriptions), Text-to-Video, Text-to-3D, and Text-to-Task (e.g., performing actions based on text input). Foundation models are large pre-trained AI models adaptable for a wide range of tasks.
Generative AI has numerous applications, including code generation (debugging, explanation, translation, documentation). Google Cloud offers services like Vertex AI Studio for exploring and customizing generative AI models, Vertex AI Agent Builder for creating chatbots and custom search engines with little to no coding, and Gemini, a multimodal AI model that understands text, images, audio, and code for complex tasks.