11 common Generative AI terms and definitions

In the previous article, we learned together what Generative AI is and how important it is in practice. However, if we just stop at the concept, it will be difficult for us to understand the nature, operation and practical application of this field. So, today let's learn more about terms related to generative AI or generative artificial intelligence!

Understand common terms in Generative AI

Common Generative AI terms and definitions

- Natural Language Processing (NLP): Considered an important branch of AI, it allows computers to understand, analyze and create human language. This is aimed at understanding and processing information in the languages ​​that people use every day more easily and effectively.

What is Natural Language Processing?

- Transformer: launched by a group of authors at Google Brain in 2017, is a neural network architecture that can train significantly larger models on larger data sets than ever before compared to the previous day. It enables the emergence of LLMs (Large Language Models - language models with language generation capabilities and natural language processing tasks) to process word sequences, recognize patterns and relationships in text. Transformer is used mainly in the fields of natural language processing (NLP) and computer vision (CV).
- Large Language Model (LLM): An in-depth model for text generation and natural language processing tasks. Nowadays almost all of these models are based on transformer architecture (with variations from the original architecture). 
For example:
+ Closed source models: GCP's PaLM, OpenAI's GPT3/3.5/4, AWS's Bedrock, Anthropic's Claude,...
+ Open source models: Google FLAN-T5 & FLAN-UL2, Falcon, MPT-7B,...
- Pre-trained Model/Foundational Model: is a machine learning model trained on a large data set and can be fine-tuned for each different specific task. Pre-trained models are often used as a starting point for developing ML models. Because they provide a set of initial weights and biases that can be fine-tuned for each specific task.

Fine-tuning

- Fine-tuning: The process of taking a model and adjusting it to suit a specific task, such as creating text, summarizing text or answering questions,... by training that model in a supervised manner. Compared to the initial model creation process, fine-tuning requires much less data and calculations but still ensures good and optimal performance.
- Hallucination: Hallucination in AI is a situation that occurs when an AI model produces a false or misleading result. And mainly the root cause is that it uses internal “knowledge” (what it has been trained on) but that knowledge is not applicable to the user's query.
- Multi-modal AI: This means that instead of one type of AI only being able to process and understand from a separate type of input, they can process different types of input such as text, words, images and videos,...
- Retrieval Model: a system used to retrieve data from the provided information source. In this case, combining retrieval models with large language models will help solve part of the illusion problem by anchoring the LLM to a known corpus.
- Vector Store: is a type of data store that specializes in managing vector representations of documents (called embeddings). These repositories are optimized to find nearest neighbors efficiently (according to various distance metrics) and are also central architectural pieces of the Generative AI platform.

Enterprise AI

- Enterprise AI: Is a form of artificial intelligence that provides users with predictive insights to improve the performance of their business processes and systems.
- Generative AI for Enterprise Search: Helps access predictive insights and underlying data of enterprise AI applications - Enterprise AI. This helps support improved decision-making for all business users.