What Is Generative AI and How Does it Work?

Generative artificial intelligence (AI) exploded on the scene in late 2022, sending people and businesses into a frenzy of curiosity and questions over its potential.

But what exactly is generative AI? Put simply, generative AI is technology that takes a set of data and uses it to create something new—like poetry, a physics explainer, an email to a client, an image, or new music.

Unlike traditional AI models, generative AI “doesn’t just classify or predict, but creates content of its own […] and, it does so with a human-like command of language,” explained Salesforce Chief Scientist Silvio Savarese.

Of course, the ability to classify and predict data accurately is a critical element to successful generative AI: The product is only as good as the data it has to work with.

There are several approaches to developing generative AI models, but one that is gaining significant traction is using pre-trained, large-language models (LLMs) to create novel content from text-based prompts. Generative AI is already helping people create everything from resumes and business plans to lines of code and digital art.

Eventually, Savarese predicts, these AI tools will “assist us in many parts of our lives, taking on the role of superpowered collaborators.” As these models become more sophisticated, they will not only enhance productivity but also foster creativity and innovation. For instance, in the creative industries, generative AI can help artists and designers explore new ideas and concepts that they might not have considered on their own. In the tech sector, developers can use these tools to generate and test code more efficiently, accelerating the development process and reducing the time to market for new products.

The integration of generative AI into everyday applications is expected to transform how we interact with technology. From personalized recommendations in ecommerce to tailored educational content, the potential applications are vast. However, this integration also brings challenges, such as the need for robust data governance to prevent bias and ensure fairness, and the requirement for transparent and explainable AI to build trust among users. As these tools become more prevalent, it will be crucial to address these challenges to ensure that the benefits of generative AI are realized without compromising ethical standards.

Digging deeper into generative AI, two primary types of deep learning models stand out: Generative adversarial networks (GANs) and transformers.

• GANs are made up of two neural networks: a generator and a discriminator. The two networks compete with each other, with the generator creating an output based on some input and the discriminator trying to determine if the output is real or fake. The generator then fine-tunes its output based on the discriminator’s feedback, and the cycle continues until it stumps the discriminator.
• Transformer models, like ChatGPT, (which stands for Chat Generative Pretrained Transformer), create outputs based on sequential data (like sentences or paragraphs) rather than individual data points. This approach helps the model efficiently process context and is why it’s used to generate or translate text.

While GANs and transformers are among the most popular generative AI models, other techniques like VAEs, autoregressive models, and diffusion models are also widely used, each with their own strengths and applications.

• Variational autoencoders (VAEs) use two neural networks: an encoder and a decoder. The encoder compresses the input data into a latent space, and the decoder reconstructs the data from this latent space. VAEs are particularly useful for generating new data that is similar to the training data but with variations.
• Autoregressive models generate outputs by predicting the next element in a sequence based on the previous elements. This approach is sequential and deterministic, making it well-suited for tasks like text generation, speech synthesis, and image generation.
• Diffusion models work by gradually adding noise to the data and then learning to reverse this process to generate new data. This method is effective for creating high-quality images and other complex data types, as it can capture intricate patterns and details.

To ensure the success of generative AI models, several key requirements must be met:

1. High-quality training data: The quality and diversity of the training data are crucial. High-quality, well-labeled data helps the model learn the underlying patterns and generate more accurate and relevant outputs.
2. Computational resources: Generative AI models, especially deep learning models like GANs and transformers, require significant computational power. Access to powerful GPUs or TPUs is essential for training these models efficiently and effectively.
3. Robust model architecture: The architecture of the model must be well-designed to handle the complexity of the task. This includes choosing the right type of model (e.g., GAN, transformer, VAE) and tuning hyperparameters to optimize performance.
4. Evaluation metrics: Robust evaluation metrics are necessary to assess the performance of generative models. Metrics like the Fréchet Inception Distance (FID) for image generation and perplexity for text generation help in quantifying the quality of generated outputs.
5. Iterative refinement: Generative AI models often require continuous testing, feedback, and fine-tuning to improve performance and address any issues that arise during training.
6. Ethical Considerations: Ensuring that the generated content is ethical and does not perpetuate biases or harmful content is critical. This involves careful curation of training data and the implementation of ethical guidelines and safeguards.

Generative AI is already making waves in various industries, with real-world applications that showcase its capabilities:

• Text Generation: Generative AI models, like GPT-3 (Generative Pre-trained Transformer 3), are capable of producing human-like text, from articles and stories to marketing content and poetry.
• Image Generation: Generative AI models, like DALL-E 2 and Stable Diffusion, have the ability to produce lifelike images from textual cues, empowering designers, artists, and content creators to realize their visions.
• Speech Generation: Generative AI models are capable of producing lifelike speech, enabling the development of voice assistants, customer service chatbots, and tailored audio experiences.

Generative AI offers a wide range of benefits that can transform businesses and industries:

• Automation of Manual Tasks: Generative AI is the key to automating repetitive and time-consuming tasks that currently require manual effort. This empowers businesses to streamline their operations, cut costs, and redeploy human capital to more strategic and creative pursuits.
• Creation of New Products and Designs: Generative AI is capable of producing entirely new products, designs, and concepts. This innovation can lead to improved customer experiences and the development of fresh solutions to complex issues.
• Enhancement of Existing Products and Designs: Generative AI can be harnessed to augment your current offerings by recognizing patterns, proposing enhancements, and optimizing performance.

Generative AI is revolutionizing the way businesses operate across various industries, driving innovation, efficiency, and new revenue streams. In the pharmaceutical industry, generative AI is being used to accelerate drug discovery. By generating and simulating new molecular structures, these models can identify potential drug candidates more quickly and cost-effectively, reducing the time and resources required for traditional drug development processes.

In the retail industry, generative AI is transforming the customer experience. Brands are using these models to create personalized product recommendations, virtual try-ons, and even generate new product designs. For example, fashion retailers can use generative AI to design unique clothing items that cater to individual customer preferences, enhancing the shopping experience and driving customer loyalty.

The entertainment industry is also seeing significant changes with the help of generative AI. Film and video production companies are using these models to generate realistic special effects, create virtual characters, and even compose music. This not only reduces production costs but also allows for more creative freedom and faster turnaround times.

In the financial sector, generative AI is being leveraged to create synthetic financial data for training and testing trading algorithms. This helps in identifying potential market trends and risks more accurately, enabling better investment decisions. Additionally, generative AI can be used to generate realistic financial scenarios for stress testing and risk management, enhancing the resilience of financial institutions.

The automotive industry is another area where generative AI is making a significant impact. Car manufacturers are using these models to design and test new vehicle components and features. By generating realistic simulations, they can accelerate the development process and reduce the need for physical prototypes, saving both time and resources.

In the technology sector, generative AI is driving advancements in natural language processing and conversational AI. Companies are using these models to develop more sophisticated chatbots and virtual assistants that can understand and respond to customer queries more effectively. This improves customer service and enhances user engagement.

Overall, generative AI is not just a technological advancement but a transformative force that is reshaping business practices and opening up new possibilities across industries. By harnessing the power of generative AI, businesses can stay ahead of the curve, innovate more rapidly, and deliver superior value to their customers.

A variety of tools and platforms are making generative AI more accessible and powerful, enabling businesses and developers to harness its capabilities for a wide range of applications.

• Agentforce (Salesforce) Agentforce is a proactive, autonomous AI application that provides specialized, always-on support to employees or customers. It can be equipped with any necessary business knowledge to execute tasks according to its specific role.
• ChatGPT (OpenAI): ChatGPT is an app powered by OpenAI’s GPT-4 LLM. It excels in generating human-like text, making it ideal for creating chatbots, customer service interactions, and content creation. Its ability to understand context and generate coherent responses has made it a popular choice for businesses looking to enhance their communication and engagement strategies.
• Gemini (Google): Gemini is a multimodal AI model from Google, designed to handle a diverse array of tasks. It can generate text, images, and even audio, making it a versatile tool for content creation and multimedia applications. Its advanced capabilities allow businesses to create rich, interactive experiences that engage users on multiple fronts.
• Claude (Anthropic): Claude is an AI model developed by Anthropic, specifically designed for friendly and safe conversations. It is built with a focus on ethical considerations and user safety, making it an excellent choice for applications where trust and reliability are paramount. Claude can be used in customer service, educational platforms, and personal assistants, ensuring that interactions are both helpful and respectful.
• Perplexity AI: Perplexity AI is an AI-powered search engine that goes beyond traditional search results. It provides direct answers to user queries and includes citations to verify the accuracy of the information. This makes it a valuable tool for researchers, students, and professionals who need quick, reliable information. Its ability to generate detailed and accurate responses enhances the user experience and saves time.
• DALL-E 3 and Midjourney: These are advanced image generation models that can create high-quality images from text prompts. DALL-E 2, developed by OpenAI, and Midjourney are widely used in graphic design, marketing, and creative industries. They allow designers to quickly generate visual content, reducing the time and effort required for manual design work and enabling more creative exploration.
• Stable Diffusion: Stable Diffusion is an open-source image generation model that has gained significant popularity due to its accessibility and flexibility. It allows developers and creators to generate high-quality images and art without the need for extensive computational resources. The open-source nature of Stable Diffusion fosters a community of innovation, with users contributing to and building upon the model's capabilities.
• Amazon Bedrock (AWS): Amazon Bedrock is a comprehensive platform provided by AWS that offers access to a wide range of foundation models. It simplifies the process of integrating generative AI into applications, making it easier for businesses to leverage these powerful tools. With services like Amazon Bedrock, businesses can quickly and efficiently incorporate generative AI capabilities into their workflows, fostering innovation and enabling new use cases.
• Amazon Q (AWS): Amazon Q is an AI assistant that is part of the developer tools provided by AWS. It assists developers in integrating generative AI into their applications, streamlining the development process and reducing the learning curve. Amazon Q developer tools provide a user-friendly interface and robust documentation, making it easier for developers to harness the power of generative AI and create sophisticated applications.

By leveraging these tools and platforms, businesses can tap into the vast potential of generative AI, driving innovation, improving efficiency, and creating new opportunities for growth and engagement.

According to Paula Goldman, Salesforce Chief Ethical and Humane Use Officer, and Kathy Baxter, Principal Architect for Salesforce’s Ethical AI practice, the potential of generative AI is enormous, but it "is not without risks."

In a coauthored article, the pair pointed out that it’s "not enough to deliver the technological capabilities of generative AI. We must prioritize responsible innovation to help guide how this transformative technology can and should be used — and ensure that our employees, partners, and customers have the tools they need to develop and use these technologies safely, accurately, and ethically."

In an interview with Silicon, Goldman shared, "Accuracy is the most important thing when applying AI in a business context because you have to make sure that if the AI is making a recommendation for a prompt, for a customer chat, or a sales-focused email, that it’s not making up facts." Ensuring data is accurate and trustworthy is foundational to any AI application. However, the training of generative AI models is data-intensive, often necessitating vast quantities of meticulously curated data, which can be a barrier for some applications.

The authoritative feel of ChatGPT responses is itself something to be mindful of, said Savarese, who warned it could lead to what he deems “confident failure.” "The poised, often professional tone these models exude when answering questions and fulfilling prompts make their hits all the more impressive, but it makes their misses downright dangerous," Saverese said. "Even experts are routinely caught off guard by their powers of persuasion." This highlights the challenge of controlling the output of generative AI models, which requires careful monitoring and intervention.

Scale the reliance on tools like ChatGPT up to the enterprise level and it’s easy to see how high the stakes could get. But IT leaders are on guard: Nearly six in 10 (59%) said they think generative AI outputs are inaccurate. This underscores the importance of addressing the issue of "hallucinations," where generative AI models, especially large language models (LLMs), can sometimes produce outputs that are factually inaccurate or nonsensical, requiring careful fact-checking.

Then there’s the question of how to use generative AI ethically, inclusively, and responsibly. The nature of generative AI raises ethical concerns related to copyright, privacy, and the spread of misinformation, requiring responsible development and deployment. Bias and fairness are also critical issues, as generative AI models can perpetuate and amplify biases present in the training data, leading to unfair or discriminatory outcomes.

That’s why Salesforce is building trusted AI capabilities with embedded guardrails and guidance to help catch potential problems before they happen. If the world is going to realize the potential of generative AI, it will need good reasons to trust these models at every level.

Responsible AI also means sustainable AI. AI consumes significantly more power than traditional workloads, and 71% of IT leaders agree generative AI would increase their carbon footprint through increased IT energy use. This highlights the computational resources required to train and run generative AI models, which can be a challenge for some organizations.

Despite the need to explore generative AI inclusively and with intention, the technology holds vast potential for the future of CRM.

The future of generative AI is exciting and full of potential. We are constantly pushing the boundaries of what is possible with new model architectures, training techniques, and applications. We believe that generative AI will have a transformative impact on many industries, changing the way we create, innovate, and interact with technology.

The societal and economic impacts of generative AI are likely to be profound, raising important questions about the future of work, creativity, and information. Responsible AI development and deployment will be crucial to ensure that generative AI is used for good and that its benefits are shared broadly.

Furthermore, the quality and management of data will remain essential. Ensuring data is accurate, updated, accessible, and complete will be vital for maximizing the effectiveness of generative AI initiatives. Companies may want to ensure that their data is harmonized to ensure accurate results.