Four Commonly Used GenAI Applications

Is the hype around Generative AI(Artificial Intelligence) worth it? Well the test of any technology is in its usefulness, in this article we will dive in depth of some common applications of Generative AI(Artificial Intelligence) and generative AI tools. But before that let us dive a little into how we ended up here, how a machine with no intelligence whatsoever started generating content.

Attention: This is where it all started

In 2017, researchers from Google Brain, including Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, and Illia Polosukhin, introduced the Transformer model in their paper "Attention is All You Need." This model revolutionized the field of natural language processing (NLP) by using a mechanism called self-attention, allowing the model to weigh the importance of different words in a sentence more effectively.

The introduction of the Transformer model marked a significant shift from previous architectures, such as recurrent neural networks (RNNs) and long short-term memory networks (LSTMs), which struggled with processing long sequences of text due to their sequential nature. The Transformer model, with its parallel processing capabilities and the self-attention mechanism, addressed these limitations and paved the way for more efficient and powerful language models.

Transformer Architecture

Not going into the specifics of it, the Transformer architecture introduced a novel approach to processing sequences of data. Unlike previous models, it relies on a mechanism called self-attention, which allows the model to weigh the importance of different words in a sentence more effectively. This architecture consists of an encoder and a decoder, each made up of layers that process the input data through multiple attention heads and feed-forward neural networks.

The encoder reads the input sequence and creates a set of continuous representations. The decoder then uses these representations, along with the output sequence, to generate predictions one step at a time. This parallel processing capability significantly improves efficiency and performance, especially with longer sequences. This method has laid the groundwork for foundation models.

Key Components

1. Self-Attention Mechanism: This component allows the model to focus on different parts of the input sequence to understand the context better. It computes a set of attention scores that indicate the relevance of each word to every other word in the sequence.
2. Positional Encoding: Since the Transformer model does not process the input sequentially, positional encodings are added to the input embeddings to give the model information about the relative positions of the words.
3. Multi-Head Attention: This involves running multiple self-attention operations in parallel, allowing the model to focus on different parts of the input simultaneously. The outputs are then concatenated and linearly transformed to produce the final result.
4. Feed-Forward Neural Networks: After the self-attention mechanism, the data is passed through fully connected feed-forward networks. Each layer in the encoder and decoder has its own feed-forward network.
5. Residual Connections and Layer Normalization: Residual connections are added around each sub-layer, followed by layer normalization, which helps in training deeper networks by mitigating the vanishing gradient problem.

BERT and GPT

Following the introduction of the Transformer, several key generative models further advanced the capabilities of Generative AI(Artificial Intelligence) and made possible many generative AI tools. One of the most notable is BERT (Bidirectional Encoder Representations from Transformers), introduced by Google in 2018. BERT's bidirectional approach allowed it to understand the context of a word based on both its left and right surroundings, leading to significant improvements in various NLP tasks.

Another groundbreaking model is OpenAI's GPT (Generative Pre-trained Transformer) series. The first version, GPT-1, released in 2018, demonstrated the potential of large-scale unsupervised pre-training followed by fine-tuning on specific tasks. GPT-2, released in 2019, significantly increased the model size and capabilities, showcasing impressive text generation and understanding. GPT-3,trained on wide range of data, released in 2020, further expanded the model's size and scope, achieving remarkable results in text generation, translation, summarization, and more.

These generative models pave the way for further development in Artificial Intelligence and generative AI tools, they are also called foundation models.

Video and Audio

Attention mechanisms have also been pivotal in the development of powerful foundation models for image and text generation and understanding also trained on wide range of data. Here are some notable examples:

DALL-E is a model developed by OpenAI is an image generator that generates images from textual descriptions using a transformer-based architecture. It can create novel and high-quality images from a wide variety of prompts.

CLIP (Contrastive Language–Image Pretraining) is another foundation model from OpenAI that learns to associate images and text by pretraining on a vast dataset of images paired with their textual descriptions. It uses a transformer-based approach to encode both modalities.

Applications of Generative AI

Let us delve into real-world applications of generative AI tools, from programming languages to virtual assistants and several range of tasks, generative AI tools has found many  business applications without human intervention.

A. Video/Image Applications

1. Video Generation

OpenAI’s Sora attracted significant attention with its impressive video generation capabilities.²

2. Video Prediction

A GAN-based video prediction system:

• Comprehends both temporal and spatial elements of a video 
• Generates the next sequence based on that knowledge (See the figure below) 
• Distinguishes between probable and non-probable sequences

GAN-based video predictions can help detect anomalies that are needed in a wide range of sectors, such as security and surveillance.

3. Image Generation

With generative AI(Artificial Intelligence), users can transform text into images and generate realistic images based on a setting, subject, style, or location that they specify. Therefore, it is possible to generate the needed visual material quickly and simply. 

It is also possible to use these visual materials for commercial purposes that make AI-generated image creation a useful element in a wide range of fields such as media, design, advertisement, marketing, education, etc. For example, an image generator, can create me writing this article.

4. Semantic Image-to-Photo Translation

Based on a semantic image or sketch, it is possible to produce a realistic version of an image. Due to its facilitative role in making diagnoses, this application is useful for the healthcare sector.

5. Image-to-Image Conversion

It involves transforming the external elements of an image, such as its color, medium, or form, while preserving its constitutive elements. 

One example of such a conversion would be turning a daylight image into a nighttime image. This type of conversion can also be used for manipulating the fundamental attributes of an image.

6. 3D Shape Generation

In this area, research is still in the making to create high-quality 3D versions of objects. Using GAN-based shape generation, better shapes can be achieved in terms of their resemblance to the original source. In addition, detailed shapes can be generated and manipulated to create the desired shape.

 B. Audio Applications

1. Text-to-Speech Generator

GANs allow the production of realistic speech audios. To achieve realistic outcomes, the discriminators serve as a trainer who accentuates, tones, and/or modulates the voice. 

The TTS generation has multiple business applications such as education, marketing, podcasting, advertisement, etc. For example, an educator can convert their lecture notes into audio materials to make them more attractive, and the same method can also be helpful to create educational materials for visually impaired people. Aside from removing the expense of voice artists and equipment, TTS also provides companies with many options in terms of language and vocal repertoire.

Using this technology, thousands of books have been converted to audiobooks.⁷

2. Speech-to-Speech Conversion

An audio-related application of generative AI(Artificial Intelligence) involves voice generation using existing voice sources. With STS conversion, voice overs can be easily and quickly created which is advantageous for industries such as gaming and film. With these tools, it is possible to generate voice overs for a documentary, a commercial, or a game without hiring a voice artist.

3. Music Generation

Generative AI(Artificial Intelligence) is also purposeful in music production. Music-generation tools can be used to generate novel musical materials for advertisements or other creative purposes. In this context, however, there remains an important obstacle to overcome, namely copyright infringement caused by the inclusion of copyrighted artwork in training data.

C. Text-based Applications

1. Idea Generation

LLM output may not be suitable to be published due to issues with hallucination, copyrights etc. However, idea generation is possibly the most common use case for text generation. Working with machines in ideation allows users to quickly scan the solution space.

It is surprising to get a machine’s help in becoming more creative as a human. This is possibly because generative AI’s(Artificial Intelligence)capabilities are quite different (e.g. more flexible, less reliable) than how we typically think about machines’ capabilities.⁸

2. Text Generation

Researchers appealed to GANs to offer alternatives to the deficiencies of the state-of-the-art ML algorithms. GANs are currently being trained to be useful in text generation as well, despite their initial use for visual purposes. Creating dialogues, headlines, or ads through generative AI(Artificial Intelligence) is commonly used in marketing, gaming, and communication industries. These tools can be used in live chat boxes for real-time conversations with customers or to create product descriptions, articles, and social media content.

Explore more large language models examples and applications like text generation.

3. Personalized content creation

It can be used to generate personalized content for individuals based on their personal preferences, interests, or memories. This content could be in the form of text, images, music, or other media, and could be used for: 

• Social media posts
• Blog articles
• Product recommendations 

Personal content creation with generative AI(Artificial Intelligence) has the potential to provide highly customized and relevant content.

4. Sentiment analysis / text classification

Sentiment analysis, which is also called opinion mining, uses natural language processing and text mining to decipher the emotional context of written materials.

Generative AI(Artificial Intelligence) can be used in sentiment analysis by generating synthetic text data that is labeled with various sentiments (e.g., positive, negative, neutral). This synthetic data can then be used to train deep learning models to perform sentiment analysis on real-world text data.

It can also be used to generate text that is specifically designed to have a certain sentiment. For example, a generative AI(Artificial Intelligence) system could be used to generate social media posts that are intentionally positive or negative in order to influence public opinion or shape the sentiment of a particular conversation.

These can be useful for mitigating the data imbalance issue for the sentiment analysis of users’ opinions in many contexts such as education, customer services, etc.

Source⁹ : “The Impact of Synthetic Text Generation for Sentiment Analysis Using GAN-based Models”

D. Code-based Applications

1. Code generation

Another application of generative AI(Artificial Intelligence) is in software development owing to its capacity to produce code without the need for manual coding. Developing code is possible through this quality not only for professionals but also for non-technical people.

Generating an HTML form and JavaScript submit code with OpenAI’s ChatGPT

2. Code completion

One of the most straightforward uses of generative AI(Artificial Intelligence) for coding is to suggest code completions as developers type. This can save time and reduce errors, especially for repetitive or tedious tasks.

3. Code review

Generative AI(Artificial Intelligence) can also be used to make the quality checks of the existing code and optimize it either by suggesting improvements or by generating alternative implementations that are more efficient or easier to read.

Author

This article was written by Zohair Badshah, a former member of our software team, and edited by our writers team.