With Artificial Intelligence (AI) evolving so quickly, it can be difficult to understand the terminology associated with it. It seems like new terms hit your feed almost weekly: NLPs, LLMs, NeRFs, Computer Vision, Deep Learning, GANs… The list goes on and on. Generative AI is one that often sparks curiosity, so let’s break it down.
The term “generative” refers to a model’s capacity to generate new content such as text, audio, images, videos, 3D assets, etc. So, Generative AI refers to a type of deep learning model that has the ability to generate content based on the data it is trained on. Think of it this way: imagine a Master Chef who has studied thousands of recipes from around the world. Thanks to the vast knowledge they’ve acquired, the chef can take a simple list of ingredients and create a completely new dish based on existing ones. A Generative AI model is like this Master Chef!
You might think of AI assistants like Apple’s Siri or Amazon’s Alexa as examples of Generative AI because they can answer your questions. However, while today’s assistants can easily be connected to generative AI models and services, it has not always been the case. They are fundamentally powered by Machine Learning technologies that perform Natural Language Processing (NLP) and Natural Language Understanding (NLU) tasks such as speech recognition and intent recognition.
Alexa and Siri extract and process user commands from natural language rather than generate content. So, if you ask them what the weather is, such assistants and chatbots extract the intent from a given utterance and compare it to a predefined weather related intent to which a data retrieval function is associated. They do not create a brand new response.
All in all, a Virtual Assistant is essentially an interface (most often voice-enabled) to interact with, and many technologies are involved in its making. As a matter of fact, NLP is still being extensively used in combination with generative AI models to perform tasks such as information retrieval which, in turn, enable models to retrieve information from documentation and use that as a basis to formulate new content.
Generative AI models work by learning from lots of data, and then making predictions. For example, models like Meta’s Llama 3 can generate text from a simple prompt. They do this by encoding the text into numbers, encoding words and characters into tokens via a tokenization process, and then decoding it back into words.
Furthermore, today’s LLMs are not limited to text as input. They have multimodal abilities which allows them to use vision and audition to respond to image and audio based prompts.
In short, a Generative AI model learns from lots of examples, and then creates “new” things; just like a chef creates new recipes. With technologies like Generative AI, platforms such as StellarX are able to create immersive and interactive experiences, making learning and working more engaging. As this technology continues to improve, we might see even more amazing creations. Understanding Generative AI helps us appreciate the exciting possibilities of what platforms like StellarX can offer!
.png){kind=small}