So far, you have learned about two paradigms of artificial intelligence, machine learning and deep learning. In this segment, we will explore the third paradigm, generative AI. We examine the tasks involved in generative AI, the nature of the training data, and the model architecture.
In the next video, Kshitij will discuss generative AI models.
Kshitij defined generative AI as follows:
“Generative AI is a subfield of AI that includes large models capable of complex generation and other tasks involving, images, music,etc.”
The most significant performance gains in generative AI models have been observed in language modelling tasks. These models can produce much longer sentences while performing sentence completion tasks and handle longer and more complex sentences while performing translation tasks.
The key technological breakthrough that facilitated these performance gains was the introduction of transformer models, In the next video, kshitij will delve into the technology behind generative models.
Kshitij discussed the transformer models in terms of the three factors we previously mentioned:
- Training data: Transformer models are-trained on billions of natural language data, made possible by the model’s larger size.
- Model size: For example, GPT-3 contains 175 billion parameters, which is orders of magnitude larger than other deep learning models. This is made possible by architectural innovations in transformer models.
- Model architecture: Transformers employ a novel architectural innovations in transformer models, as outlined in the image given below.
He then highlighted the advantages resulting from the architectural innovations in transformer models, as outlined in the image given below.
Before moving forward, evaluate your understanding of the topics covered in this segment by answering the following questions.
Finally, Kshitij will explain the practical implications of these performance gains in terms of the applications of large language models.
In the video, you learned that one large language model can be used for various language tasks. Moreover, generative models are not limited to language tasks only; they can also be utilised for generative tasks involving images, videos, audio, and more. Kshititij discussed the tasks, training data, and models associated with generative AI models. The table below summarises this information for the three types of AI models we discussed.
You have now learned about the different paradigms of artificial intelligence, with a particular focus on generative AI This concludes your journey into generative AI. In the next segment, we will summarise what you learned this session.