Transformers, transformers everywhere: An Overview of Transformers

Many artificial intelligence enthusiasts and professionals might agree that the capacities of recent deep learning models have made leaps and bounds. The recent unveiling of models such as OpenAI's successor for DALL-E, DALL-E 2 [1] and Google's Imagen [2] have shocked the world with completely artificial images based on text prompts. Though for the greater public, the graphical aspect of these models catches the eye easier, the high levels of Natural Language Understanding (NLU) shown in these models is enough to impress anyone. You can also generate the coolest dog you have ever seen, which is obviously the real breakthrough here. These models address many tasks, including but not limited to: language understanding, question-and-answer for customer service chatbots, text classification for spam and fraud detection, text completion and generation for assisted content creation, image captioning and image segment captioning for automatic labeling and automatic image description, sentiment analysis for hate speech detection, style transfer…These tasks are highly relevant as information flow across web platforms keeps growing larger and larger and state-of-the-art models are needed to keep up.