Upon its release in November 2022, ChatGPT stunned Silicon Valley and the world. OpenAI, a small company based in San Francisco, introduced a chatbot that mimics complex emotions, writes code and answers complex questions.
Technology considered a decade away was now at everyone’s fingertips and quickly became the fastest-growing app in history. Just four months later, OpenAI launched a significant update, ChatGPT-4, and the results of this new technology are fascinating:
- ChatGPT tricked a human into solving a Captcha test for it, telling the person hired on TaskRabbit that it was blind.
- A judge in Colombia used the tool to assist in ruling on a medical rights case.
- It scored in the top 10% in the Uniform Bar exam and passed a U.S. medical licensing exam.
ChatGPT-4 is the latest and most exciting development of generative AI, but it’s far from the only tool transforming our approach to AI. DALL-E creates impressive art on demand, such as a medieval painting of a broken Wi-Fi connection. (The name is a mash-up of the famous artist Salvador Dalí and the beloved cartoon robot WALL-E.) Other models create business simulations, videos, code and audio. Generative AI has transformed industries and reshaped how we as a society approach technology.
What is there to know about the latest evolution in technology? Let’s dive into generative AI: what it is, how it works and its exciting applications in the world.
What is Generative AI?
Generative AI is artificial intelligence (AI) that creates different types of content, such as text, images, audio, videos and 3D models. While traditional AI focuses on identifying patterns, polishing analytics, making decisions, and detecting fraud, generative AI focuses on:
- Learning patterns from existing data.
- Using those patterns to generate realistic and unique outputs.
At the core of generative AI are generative models. These models are trained on large datasets and learn to generate new content by understanding the underlying structure and relationships within the data. The training process typically involves optimizing the model’s parameters to minimize the difference between the generated outputs and the actual data — allowing the model to improve its ability to create realistic and coherent content.
(Read about adaptive AI, which is earlier in the AI hype cycle.)
How Generative AI works
Generative AI isn’t new. It was first developed in the 1960s in the form of chatbots. The machine learning used in generative AI was mainly limited to predictive models and only used in observing and classifying patterns. However, generative AI broke through the traditional barriers and limitations of machine learning. It was no longer limited to perceiving and organizing information: it could now create an image or text description on demand.
There are a few types of generative AI models, and they work in different ways. The three most popular ones are GABs, VAEs and transformer-based models, like GPT-4.
Generative Adversarial Networks (GANs)
In 2014, Ian Goodfellow introduced the GANs model to the world. It took the tech world by storm because it was the first to create realistic-looking and sounding people.
These models work like a game between two players. One player (the generator) creates the new content, while the other (the discriminator) tries to determine whether it’s real or fake. As the generator gets better at creating realistic content, the discriminator improves in identifying it. They both learn together to create more realistic new content.
Variational Autoencoders (VAEs)
Around the same time, Goodfellow came up with GANs, Diederik P. Kingma and Max Welling introduced variational autoencoder (VAE), an artificial neural network, in 2013. It learns to represent and generate data probabilistically and consists of two components: an encoder and a decoder.
The models work like an artist who first makes a simple sketch, then adds details to create a new picture. VAEs learn to simplify examples they see and then use the simplified version to create new, unique content.
Transformer-based models (like ChatGPT-4)
Transformers are the machine learning model that made the most recent advancements in generative AI possible because it doesn’t require labeling the data in advance. It enables the technology to train with billions of pages of data and text, which results in more depth.
Transformers also enabled the concept of attention, allowing the technology to track connections across pages, chapters, and books instead of individual sentences. This works for more than just words, too: it can track connections to analyze chemicals, proteins, codes and DNA.
The advancements of large language models (where models have billions or possibly trillions of parameters) have created the next era in generative AI that we see today.
Benefits of Generative AI
The latest enhancements in generative AI have incredible implications and applications for society and business today. Some organizations have already started implementing generative AI initiatives, including developing custom ones with proprietary data. Some of the most significant benefits of generative AI include the following:
While many content creators and artists have voiced worries about being replaced by AI, the technology’s capabilities are not there yet. Instead, it’s a valuable tool to work alongside artists, designers, writers and musicians to explore new creative avenues.
Together, they can generate unique and inspiring content based on existing patterns and styles. It can be a creative partner to help individuals push their imaginations’ boundaries.
Prolific content generation
Generative AI can speed up the content generation process by automatically generating drafts, designs or compositions that professionals can refine and improve. It helps save time and resources, allowing creators to focus on perfecting the final product.
Customization and personalization
Generative AI creates personalized content tailored to individual preferences, such as:
- Custom artwork
- Marketing materials
- Product recommendations
Its ability to enhance personalization can lead to better customer experiences and increased customer satisfaction.
In fields like machine learning and computer vision, generative AI can create additional training data by generating new examples that share characteristics with the original dataset. This can lead to better model performance, particularly when limited available data exists.
Simulations and prototyping
Generative AI can quickly generate multiple design prototypes, which enables engineers and designers to explore different solutions and make informed decisions. Plus, generative models can simulate real-world scenarios, which help researchers and policymakers test various strategies and approaches.
Education and training
Students and educators can use generative AI to create customized learning materials and resources to meet unique needs and learning styles. It can also be used to develop training simulations, enabling professionals to practice and improve their skills in a safe, virtual environment.
Risk & reward
While generative AI offers significant benefits, using the technology responsibly is essential. It does present potential ethical concerns, such as bias, misuse and job displacement. By addressing these challenges, generative AI can continue positively transforming industries and creative fields.
(Consider what generative AIs mean for cybersecurity: it’s risk & reward.)
Pioneering the future of AI
Generative AI is redefining the technology world and dramatically transforming how companies and people learn concepts, create content, purchase, and use future products.
However, it needs to be stressed that the level of generative AI we’re seeing develop is relatively new. The risks and opportunities it presents will likely evolve in the coming weeks, months and years. As it becomes increasingly integrated into society and business, it will give us new benefits, risks — and a regulatory climate in response.
As companies start to experiment and find value with generative AI tools, leaders need to keep tabs on regulations, use cases and potential risks.
What is Splunk?
This posting does not necessarily represent Splunk's position, strategies or opinion.
Common Event Format (CEF): An IntroductionTop DevOps Certifications to Earn in 2023Best Cybersecurity Certifications for Pros to Earn in 2023Splunk Threat Intelligence ManagementSupply Chain Attacks: What You Need to Know