Primer on Generative AI

What is Generative AI?

Generative AI is a type of artificial intelligence that is able to create diverse content (e.g., text, images, video, audio, etc.).

How does Generative AI work?

Models are trained to generate novel output by "predicting" the next element in a sequence based on the context of preceding elements, allowing them to generate coherent and contextually relevant content.

What is a LLM?

LLM is an abbreviation for Large Language Model. This type of model, such as GPT-4 from OpenAI, uses machine learning to generate human-like output. LLMs are trained on vast amounts of data to understand and generate human-like language in order to make "predictions" of what is the next element in a sequence based on preceding elements. LLMs are called "large" because they have a large number of parameters. For example, the GPT-4, which is currently one of the biggest models, has 1.8 trillion parameters. These vast numbers of parameters permit the models to grasp an extensive variety of language patterns and subtleties.

What kind of applications can I make with Generative AI?

There is a wide array of potential applications for Generative AI models across many industry sectors. Some examples include:

  1. Content Generation:

    • Text Generation: Generate articles, stories, poetry, or other written content.

    • Code Generation: Automatically generate code snippets based on a given description or task.

    • Image Generation: Create unique artworks, designs, or illustrations.

    • Video Synthesis: Generate new videos or modify existing ones.

    • Voice Generation: Create realistic synthetic voices for various applications, including voiceovers and voice assistants.

  2. Conversational Agents:

    • Chatbots: Develop conversational agents that can engage in natural language conversations.

    • Copilots: assistants that can guide humans in achieving goals / outcomes.

  3. Analysis / Synthesis:

    1. Data Anomalies: Detect anomalies in data by generating expected patterns and identifying deviations.

    2. Simulated Environments: Generate realistic simulations for training purposes in various fields, such as autonomous vehicles or medical training.

  4. Vertical use cases:

    1. Game Design: Generate game levels, characters, and other content dynamically.

    2. Molecule Generation: Assist in generating new molecular structures for drug discovery.

What is an embedding?

Embeddings are the numerical representations of concepts, which are translated into sequences of numbers, making it easier for computers to comprehend the relations between those concepts. They can capture the context of a word within a document, as well as its semantic and syntactic similarity and its relationship with other words.

What is a VectorStore?

A vector store refers to a storage system or a database that can efficiently handle vector data. Vector data is when high-dimensional vectors are utilized to represent complex data like words, phrases or images.

What is the memory of an LLM?

An LLM’s memory refers the extent of which the model can consider previous text and responses when generating new text. The model’s “memory” is significant when dealing with long dialogue or text, as it determines the extent of prior context the model can use to produce precise and insightful responses.

What is a multimodal model?

A multimodal model in the context of artificial intelligence signifies a model capable of managing and integrating data from several different types of input or modalities. These can encompass text, images, sound, video, and more. For instance, a model that uses both text and image as input may have a better understanding of the context than a model only utilizing one of them.

Last updated