Unleashing Productivity: Dive into Generative AI’s Game-Changing Industry Applications

16 min readMay 1, 2024

Here are the links for the previous parts of the blog series:

Part 1: Unleashing Creativity: An Introduction to Generative AI | by NeuroCortex.AI | Feb, 2024 | Medium
Part2: Unraveling the Magic: A Deep Dive into Generative AI | by NeuroCortex.AI | Mar, 2024 | Medium
Part 3: Implementing Generative AI: A Pipeline Architecture | by NeuroCortex.AI | Apr, 2024 | Medium

A steampunk woman reading a book with her cat at the bar. Image generated by Stable Diffusion

Imagine a world where artificial intelligence can not only analyze data but also create entirely new things. This is the exciting realm of generative AI, a rapidly evolving field with the potential to revolutionize countless industries.

Generative AI goes beyond simple automation. It harnesses the power of machine learning to produce entirely new content, from realistic images and captivating music to innovative product designs and engaging marketing copy. This technology is no longer science fiction; it’s actively transforming how we work and create.

Few results from OpenAI’s SORA text to video model

In this captivating exploration, we’ll embark on a journey to unveil the multifaceted applications and delve into the fascinating world of generative AI. We’ll witness how it’s transforming industries and uncover its diverse applications across various sectors, igniting a spark of innovation across diverse fields, explore its core concepts and discuss the real-world impact it’s already having.

Here we will discuss about challenges faced by a fellow Gen AI practitioner named Monique; she was a Technical Program Manager with a vision to develop an unique tool for clients to perform automatic text to sql queries. She was primarily solving natural language to SQL problem using GPT 3.5 version.

The key goal is to create a tool that allows non-technical users to easily present their work without needing to write SQL queries directly. Instead, users can input their requests in plain English, and the tool will generate the necessary SQL commands to retrieve the desired results. This tool will cater specifically to the organization’s database, which may contain normalized data and complex problem statements.

Many existing text-to-SQL tools struggle with normalized data and complex queries. Monique’s custom solution is supposed to address these challenges by incorporating custom logic to handle complex queries and normalized data effectively. Additionally, the tool will have the capability to make API calls to access data stored in other databases and formats. It will be able to handle not only tabular data but also unstructured data, which is common in many organizations. This custom-made solution aims to empower users to quickly and effectively present their work in conferences and meetings without requiring technical expertise in SQL.

The data may also be in PDF format or stored in documents located elsewhere, not necessarily confined to databases like MongoDB or any other NoSQL databases. The user should have the flexibility to store data in various locations such as blob storage.

The initial step involves intent classification to discern the user’s objective, such as determining the type of data they wish to access. AWS offers services for intent classification, and RASA provides feedback loops for intent classification, employing models like Neural Nets.

The main issues Monique faced during validating output:

GPT initially struggled to distinguish whether user was referring to a column or a condition in a query. For instance, if user requested the median income for the region “Africa” from a table containing a “region” column with different country names, GPT might interpret it as asking for the mean of a column named “Africa.” To address this issue, her team had to compile a large dataset of relevant text-to-query examples. Initially, they had a limited number of records, assuming that GPT, as a few-shot learner, would perform adequately. However, soon she found it necessary to explicitly clarify in the prompt that user intends to filter data based on a condition rather than selecting a column.
Results were not consistent. It used to give different result each time. So she found it tough to give demo to their clients.
Hallucination — On many occasions, rather than generating SQL queries, GPT provided Python code. To address this issue, her team adjusted the prompt. They instructed the system that if it was unsure about the SQL query, it should indicate so and refrain from producing any result.
Another approach she tried was to give data to GPT and ask it to directly give the summary statistics instead of giving a query. Results were wrong for more than 60% times.
Suppose we have a table with two columns: “cost” and “maximum cost.” If you request the system to provide the maximum cost, there might be an issue with intent classification. The system may not discern whether you are asking for the maximum value within the “cost” column or referring to the entire “maximum cost” column. Addressing this ambiguity solely through NLP can be challenging. (To resolve this, we can prompt users to clarify the ambiguity. We can ask them whether they are seeking the maximum value of the “cost” column or the entire “maximum cost” column. Utilizing IF-ELSE conditions becomes necessary in such cases. Relying solely on the model may not suffice; human intervention and clarification are often required. We should aim for more than 80% accuracy in handling such ambiguities, as 60–70% accuracy would not be satisfactory)
The human-in-the-loop approach involves verifying the accuracy of the response output. Initially, feedback can be requested from the client, asking if they are satisfied with the provided response. Clients can then indicate their satisfaction level through actions such as giving a thumbs-up or thumbs-down or providing a rating to validate the output.
Periodically, a team member samples some responses and evaluates them to assess the system’s performance. If the responses are accurate, they are included as part of the training data. Subsequently, the model undergoes fine-tuning to enhance its performance further.

As of this writing the above said tool is still being built and continuously improved in order to become industry ready and consistent.

**Venn Diagram of Artificial Intelligence, ML, DL and GenAI techniques**

The Timeline of Various Gen AI Techs

Timeline showcasing all the major developments in the field on GenAI

The Emerging Generative AI Tech Stack

The generative AI tech stack is a comprehensive breakdown of the tools, technologies, and frameworks commonly deployed in the development of generative AI systems.

It comprises three fundamental layers: the application layer, the model layer, and the infrastructure layer. The application layer includes end-to-end apps or third-party APIs that integrate generative AI models into user-facing products. The model layer comprises proprietary APIs or open-source checkpoints that power AI products, requiring a hosting solution for deployment. The infrastructure layer encompasses cloud platforms and hardware manufacturers responsible for running training and inference workloads for generative AI models.

Gen AI Tech Stack highlighting Application, Model and Infrastructure layers.

The AI tech stack is crucial for building a reliable and effective generative AI system. It simplifies the development process, accelerates innovation, and enables faster development and deployment of generative AI applications. The tech stack comprises various components, such as machine learning frameworks (TensorFlow, PyTorch, Keras), programming languages (Python, Julia, R), data preprocessing tools (NumPy, Pandas, OpenCV), visualization tools (Matplotlib, Seaborn, Plotly), and deployment tools (Flask, Docker, Kubernetes).

Artistic visualization of AI pipeline with human in the loop (Picture credits: abacus.ai)

Generative AI, unlike other forms of AI, requires a unique tech stack to create new content, whether it’s text, images, or even code. This stack can be broken down into several layers, each playing a crucial role in the development process.

Gen AI Tech Stack with descriptions of each layer

Foundational Layers

Deep Learning Frameworks: TensorFlow and PyTorch are two popular frameworks that provide the infrastructure for building and training complex models. These frameworks allow for the implementation of different generative AI architectures, including Generative Adversarial Networks (GANs), Recurrent Neural Networks (RNNs), and Variational Autoencoders (VAEs).
Programming Languages: Python is the most commonly used language for generative AI development due to its extensive libraries for scientific computing and data manipulation, like NumPy, Pandas, and Matplotlib. Additionally, Julia is gaining traction due to its focus on speed and efficiency.

Generative Model Techniques

Foundation Models (FMs): FMs are pre-trained machine learning models that act as the backbone of generative AI systems. These models are trained on massive datasets and can be fine-tuned for specific tasks like text generation or image creation. Hugging face has a complete repository of open sourced models

Hardware

Compute Hardware: Training generative models necessitates significant processing power. GPUs (Graphics Processing Units) and TPUs (Tensor Processing Units) are specialized processors that can accelerate the training process significantly. These processors are often used in conjunction with high-capacity storage systems to manage the massive datasets involved in generative AI. For example Meta used 2048 x Nvidia A100 GPUs to train LLaMa model.

Deployment

Cloud Platforms: Cloud platforms like Google Cloud Platform (GCP), Amazon Web Services (AWS), and Microsoft Azure offer scalable infrastructure and resources for deploying and managing generative AI models.

The Foundation: Deep Learning Frameworks and Languages

Imagine the generative AI tech stack as a pyramid. At the base are deep learning frameworks like TensorFlow and PyTorch. These frameworks provide the building blocks for crafting complex models, the engines that power generative AI. They allow you to implement various architectures like Generative Adversarial Networks (GANs), renowned for creating realistic images, and Recurrent Neural Networks (RNNs), which excel at text generation.

The Heart: Generative Model Techniques

Atop the foundation lies the heart of generative AI — generative models. Here’s where the magic truly happens. Foundation Models (FMs) are pre-trained machine learning models that serve as the backbone of generative systems. Pre-trained on massive datasets, these models can be fine-tuned for specific tasks, allowing them to generate text, translate languages, or create images tailored to your needs.

Flow of information from the hardware in the cloud to the end users. Different colors highlight which layer of Gen AI they belong to.

As already mentioned generative AI tech stack comprises three fundamental layers:

The applications layer includes end-to-end apps or third-party APIs that integrate generative AI models into user-facing products.
The model layer comprises proprietary APIs or open-source checkpoints that power AI products. This layer requires a hosting solution for deployment.
The infrastructure layer encompasses cloud platforms and hardware manufacturers responsible for running training and inference workloads for generative AI models.

Generative AI tech stack with industry recognized software solutions

Generative AI Use Cases

Prominent Gen AI tools in domains of text, code, images/video and audio

Generative AI has found applications across various domains, enabling creative and innovative solutions. Here are some prominent use cases:

Text Generation:

Content Creation: Generating articles, blog posts, and other written content.
Chatbots: Creating natural language interfaces for customer support or general conversation.
Code Generation: Generating code snippets or even complete programs based on high-level descriptions.

Text and code

a) GPT-4 on chatGPT (essays,blogs,articles,Q/A)
b) AlphaCode(code)
c) GitHub Copilot(code)
d) BARD(essays,blogs,articles,Q/A)
e) Jasper.ai(Marketing content)
f) Writesonic(content creation)
g) Copy.ai(content creation)

Image Generation:

Art and Design: Creating digital art, illustrations, and graphic designs.
Photorealistic Rendering: Generating realistic images for architectural visualization, product design, or virtual environments.
Style Transfer: Applying artistic styles to images.

Images (Text to Image)

a) DALL-E 2
b) Midjourney
c) VQGAN+CLIP
d) Starry Night (can create painting in the style of famous artists)
e) DeepDream
f) Artbreeder
g) Leonardo.Ai

Video Generation:

Deepfake Technology: Creating realistic videos by altering facial expressions or replacing actors.
Video Synthesis: Generating video content based on textual or visual input.

Video generation and editing

a) Descript
b) Wondershare Filmora
c) Runway
d) Peech
e) Synthesia
f) Fliki
g) InVideo

Music Composition:

Music Generation: Composing original music or generating music in specific styles.
Lyric Generation: Creating lyrics for songs.

Audio

a) Lovo.ai
b) Syntheses
c) Murf
d) Voice Over by Speechify
e) Altered
f) Listnr
g) Speechelo

Music

a) Amper Music
b) MuseNet
c) Soundraw
d) Melobytes
e) Boomy
f) Magenta
g) AIVA

Game Development:

Procedural Content Generation: Generating game levels, environments, and characters.
Dialogue Generation: Creating dynamic and context-aware in-game dialogue.

Here are some resources to explore further:

A curated list of AI tools for game developers: [GitHub — simoninithomas/awesome-ai-tools-for-game-dev: A curated list of awesome AI tools for game developers]
Top AI tools for game development: [Top 10 AI Tools Transforming Game Development — Analytics Vidhya]

Language Translation:

Neural Machine Translation: Improving the accuracy and fluency of translations between languages.

Here are some resources to explore further:

The Top 10 AI Language Translation Tools to Try in 2024 | TW Languages

9 AI translation tools you need to try in 2024 — Lokalise Blog

Data Augmentation:

Data Synthesis: Generating synthetic data for training machine learning models, especially in scenarios with limited real-world data.

Here are some resources to explore further:

Real-Time Tools for Generative AI, Data Augmentation, and Live Data Streaming | LinkedIn

Drug Discovery:

Molecule Generation: Designing new molecules for drug development.
Chemical Reaction Prediction: Predicting possible reactions between different chemical compounds.

Popular Gen AI Drug Discovery Tools

AIDDISON™ by Merck KGaA:This AI-powered platform integrates various AI methods, including generative modeling, for virtual screening, lead discovery, and lead optimization.
BenevolentAI: This company utilizes gen AI to design new drugs and identify new uses for existing medications.

Top 6 AI-Powered Drug Discovery Tools In 2021 (analyticsindiamag.com)

Storytelling:

Plot and Dialogue Generation: Assisting writers in developing storylines and character dialogues.

Popular Storytelling Gen AI Tools

Jasper: This AI writing assistant offers a suite of features for storytellers, including scene generation, dialogue writing, and character development. [Jasper AI]
Sudowrite: This AI tool focuses on long-form writing and provides in-depth story development support, character building, and plot refinement. [Sudowrite]
NovelAI: This immersive platform allows you to train the AI on specific writing styles or genres, making it ideal for crafting stories in a particular voice. Additionally, it can generate images to depict your characters. [NovelAI]

5 Best AI Story Generators in 2024 (Ranked) (elegantthemes.com)

Top AI Storytelling Tools | AI Scout

Medical Imaging:

Image Synthesis: Generating synthetic medical images for training and testing diagnostic models.

Popular Gen AI Tools in Medical Imaging

Blackford:This AI-powered platform utilizes generative models to reconstruct medical images, particularly CT scans, reducing noise and enhancing details for improved analysis.
Viz.ai:This tool leverages AI to analyze chest X-rays, streamlining the workflow for radiologists and potentially expediting the detection of critical conditions like pneumonia.

Top AI Medical Imaging tools for Healthcare Analytics — 2024 | Factspan

Top AI based Medical Imaging Tools In 2021 | Factspan

Speech Synthesis:

Voice Generation: Creating synthetic voices for virtual assistants, audiobooks, or accessibility applications.

Top Gen AI Speech Synthesis Tools:

ElevenLabs: This platform offers high-quality text-to-speech with a wide variety of voices, including the ability to clone your own voice. [ElevenLabs]
Synthesia: Synthesia provides a vast library of AI voices in over 120 languages, allowing you to generate realistic voiceovers for your videos. They also offer voice cloning capabilities. [Synthesia]
Respeecher: Aimed at media productions, Respeecher excels at high-fidelity voice cloning. It allows for detailed voice customization to perfectly match the target voice. [Respeecher]
Lovo.ai: This user-friendly tool offers ultra-realistic voices, along with features like custom pronunciation rules and support for over 100 languages. [Lovo.ai]
Replica Studios: Known for its intuitive interface, Replica Studios provides multiple export formats and streamlines script and line management for efficient audio creation. [Replica Studios]

13 Best AI Voice Generators (Text-to-Speech) of 2024 (synthesia.io)

5 Best AI Voice Generators: AI Text-To-Speech In 2024 (eweek.com)

Personalization:

Recommendation Systems: Generating personalized recommendations for products, content, or services.

Popular Personalization Gen AI Tools

Adobe Target: This AI-powered platform offers A/B testing, personalization across channels (website, mobile app, etc.), and real-time optimization based on customer behavior. [Adobe Target]
Insider: This comprehensive tool provides features like AI-powered personalization, customer segmentation, and recommendation engines, allowing businesses to tailor experiences across all touchpoints. [useinsider.com]
Persado: This AI platform focuses on creating personalized marketing copy that resonates with different audience segments, helping businesses improve marketing campaign effectiveness. [persado.com]

Simulation and Training:

Simulated Environments: Generating virtual environments for training autonomous vehicles, robots, or simulations for various industries.

Popular Gen AI Simulation and Training Tools

SimulateMe: This platform leverages AI to create immersive virtual reality (VR) training simulations for various industries, from soft skills training to healthcare procedures. [SimulateMe]
Mursion: This AI-powered platform focuses on language learning and cultural awareness training, using virtual simulations to create realistic conversations and scenarios in different languages and cultural contexts. [Mursion]
VictoryXR: Specializing in military and defense training, VictoryXR utilizes AI to generate complex and adaptive combat simulations for soldiers to hone their decision-making and tactical skills in a virtual environment. [VictoryXR]

Finance and Trading:

Market Prediction: Analyzing historical data to generate predictions and insights for financial markets.

Popular Gen AI Tools for Finance and Trading

For Individual Investors:

Earni: This platform utilizes AI to analyze your financial situation and investment goals, recommending personalized investment strategies and helping you build a diversified portfolio.
Quantiacs: This platform allows users to develop, test, and deploy automated trading algorithms using gen AI and machine learning tools.

For Financial Institutions:

Kensho (kensho.com): This AI-powered financial analytics platform helps professionals gain insights from complex financial data, facilitating informed investment decisions.
AlphaSense (alphasense.com): This tool uses AI to process vast amounts of financial news and information, allowing analysts to identify relevant trends and make data-driven investment decisions.

10 AI Tools for Stock Trading & Price Predictions — GeeksforGeeks

These use cases showcase the versatility of generative AI across different industries, highlighting its potential to drive innovation and enhance various processes.

More Examples of Gen AI Applications by Use Cases (Created by Generative AI: Creating machines more human-like (kaggle.com))

Generative AI Tools

Below are different tools that can be used for different output generation.

Industry-Specific Applications

Challenges

Generative AI models may consist of billions of parameters, and their training methods require quick and effective pipelines of information. A sizable amount of financial investment, technical know-how, and computing infrastructure are required to construct and maintain generative models.
Conversations must occur quickly and precisely for collaborative use cases like chatbots, AI voice assistants, or customer service applications. Diffusion models’ slow sampling rates have come to light as a result of their growing popularity and the high-quality samples they can provide.
To create artificial data for various use cases, generative AI models are frequently utilized. Still, not all data can be used to train AI models, even though enormous amounts of data are produced daily around the globe. To function, generative models need reliable, unbiased information.
Getting commercial authorization to use pre-existing datasets or to create new evidence-based datasets to train generative models is a challenge for many organizations. This procedure is crucial for preventing concerns about intellectual property violations.
Regarding the use of creative AI tools, legal and ethical issues can arise. One of these abilities is the ability to quickly produce “deepfakes,” which are computer-generated images or movies that appear genuine but are fake or misleading. ChatGPT is already causing problems for universities. They are particularly having trouble with generated degree work since it is simpler to duplicate. Midjourney and Dall-E have faced similar criticism from artists who view them as either a form of stealing or just “not art.”
Machine learning, statistics, and computer science are just a few of the highly specialized fields that must be thoroughly understood to work in the field of generative AI. This may make it difficult for a small number of businesses to design and execute unique generative AI solutions, which may slow down the use of generative AI and hike compliance costs. Additionally, there is a significant demand for skilled individuals in the field of generative AI, but only a few are available.
The next difficulty is assuring generative AI’s social and human effect, including preserving trust and openness, encouraging inclusiveness and diversity, and promoting ethical and advantageous use cases.
Making sure that generative AI is used in a way that respects intellectual property rights, protects security and privacy, and steers clear of dangerous or immoral use cases is another challenge.

Summary

This blog series explored the exciting potential of generative AI, a type of artificial intelligence that can create entirely new content. We journeyed beyond the realm of data analysis and witnessed how generative AI can revolutionize industries through its ability to generate:

Creative Content: From music and images to marketing copy, generative AI can produce high-quality content, boosting efficiency and innovation.
Novel Designs: Generative AI can design new products, optimize processes, and even craft virtual worlds, pushing the boundaries of what’s possible.
Enhanced Data: By creating synthetic data, generative AI can improve the accuracy and effectiveness of machine learning models.

These are just a few examples of how generative AI is making waves. As the technology matures, we can expect even more transformative applications across various sectors. Stay tuned for future posts where we’ll dive deeper into specific use cases and explore the exciting future of generative AI.

Conclusion

Generative AI has unveiled a future brimming with creative and practical possibilities. We’ve explored its remarkable ability to generate new content, designs, and data, showcasing its transformative potential across industries.

However, the journey of generative AI is just beginning. As the technology continues to evolve, we can expect even more groundbreaking applications to emerge. Here are some key takeaways to consider:

The Human Touch Remains Crucial: Generative AI is a powerful tool, but it shouldn’t replace human creativity entirely. The best results will likely come from a collaborative approach, where AI augments and inspires human ingenuity.
Ethical Considerations: As with any powerful technology, ethical considerations around bias and responsible use are paramount. Open discussions and robust guidelines are necessary to ensure generative AI is used for good.
A Future Full of Potential: The potential applications of generative AI seem limitless. From personalized experiences to accelerated scientific discovery, this technology holds the key to unlocking a brighter future.

We stand on the precipice of a generative revolution. By embracing this technology thoughtfully and responsibly, we can unlock a world brimming with innovation and progress. Stay curious, stay informed, and get ready to witness the incredible ways generative AI will continue to shape our world.

We give special thanks to Abhijeet Upadhay for providing us with detailed feedback and sharing his expertise on the industry use-case (text2sql) we presented in this blog.