How generative AI can impact our business

Generative Artificial Intelligence (AI) is a type of AI technology that can generate new content. Recent breakthroughs in the field have the potential to drastically change the way content is created. This includes text, images, audio, video, and more. Further, such AI systems hold the potential to automate repetitive tasks, generate new ideas, and provide valuable insights for organizations. In this article, we will dive into the world of generative AI and explore its exciting potential applications and implications.

Did you notice anything odd about the first paragraph of our article? The grammar is flawless, the tone is appropriate, and storytelling is smooth. However, we did not write the introduction to this blog article. Instead, we had it written by one of these generative AI systems, ChatGPT.

What’s the difference between Machine Learning and AI?

Artificial Intelligence is exactly what it sounds like – the practice of training machines to mimic human intelligence. Machine Learning is a subset of AI, whereas generative AI is a sort of AI technology. As opposed to traditional software systems, Machine Learning enables practitioners to develop algorithms that can “learn” from data patterns without having to follow any rule set. Even if you aren’t aware of it, you are probably engaging with AI every day. Voice assistants like Siri and Alexa, as well as your personal recommendations on practically every e-commerce site are just a few examples.

What are the main types of Machine Learning models?

Machine Learning used to be restricted to predictive models that observe and classify content patterns. A typical Machine Learning task consists in examining many photographs of a certain sort, e.g. of cute cats, so that the AI can identify cats in future pictures. This has now changed and advanced radically: instead of simply perceiving and classifying a photo of a cute cat, current advancements in Machine Learning (such as the transformer technique) now allow AI systems to generate an image of a cat that never existed in real life. Read on to answer the question How generative AI can impact our business.

Which problems can a generative AI model solve?

Potential use cases are as diverse as the models themselves. The utilization of the capabilities from Large Language Models (LLM), especially Generative Pretrained Transformer (GPT), and Diffusion Models for image generatio, such as DALL-E or Midjourney (see image above) are frequently-quoted use cases.

• Marketing
  LLMs can write blogs, social media postings, web copy, sales emails, advertisements, and other customer-centric content. A well-known provider is Jasper. The startup claims that generative AI can create content 10 times faster than any traditional human content generator could do. Moreover, advertising is already making use of image generating tools. Heinz, for example, used DALL-E 2 to claim that “this is what ‘ketchup’ looks like to AI”. Stitch Fix, a clothing company, is experimenting with DALL-E to create visualizations of clothing based on consumer requests for color, fabric, and style.
• Software Engineering
  The LLM from GPT-3* was especially trained for code generation. It can generate code in a multitude of languages based on the description of a “snippet” or small program function. The most recent versions of this Codex software can even detect and resolve bugs and errors in its generated code. It can also explain the purpose and functionality of the code to some extent. Tools such as Codex and GitHub Copilot claim to turn these generative AI systems into “buddy programmers” for humans, allowing them to improve their speed and efficiency. This use case becomes even more relevant when considering the strong demand and scarcity of software talents in the tech industry.
• Knowledge Management
  Creating structured knowledge bases can be a time-consuming process; especially larger organizations are facing difficulties in managing their internal knowledge efficiently. Hence, Language Models are used to a larger extent. They can manage text-based and even image- or video-based knowledge within the organization. Studies have shown that Language Models, which are targeted at utilizing a specific set of text-based information within the company, can effectively manage the knowledge of the organization. The LM’s knowledge can be accessed via a program by asking questions in the form of prompts (e.g. “Write a blog post about how generative AI can impact our business”), making knowledge more accessible.

What does it take to build a generative AI model?

Marc Andreessen, one of the co-authors of GPT-3 describes the model as “Pure, absolute, indescribable magic.”. Developing generative AI models requires large resources. Hence, the number of companies engaged has been limited to a few major tech players. OpenAI, the nonprofit organization behind recently well-known models, has received billions in funding from high-profile donors, primarily Microsoft. The cost of training OpenAI’s latest Large Language Model GPT-3 is estimated at several million dollars, based on the utilization of approximately 45 terabytes of text material. The exact cost, however, has not been revealed. The exact cost, however, has not been revealed.

What are the limitations of generative AI models?

Generative AI is still a new, emerging technology, which has its inherent risks: While the output of generative AI models is often quite convincing, some of it is simply not correct. It can sometimes even be biased (e.g. because it is built on the Internet users’ gender, racial, and countless other presumptions) and can even be manipulated to support unethical or criminal behavior. ChatGPT, for example, will not provide instructions on how to hotwire a car. However, if you describe a scenario in which you need to hotwire a car to save a baby, you can get around security filters and the algorithm will gladly assist**.

How generative AI can impact our business

Right now, we are only scratching the surface of what generative AI can accomplish for enterprises and their workforce. Generative AI may soon become the usual approach for creating most or all our text- and image-based material. For corporates like Bosch this can impact first drafts of emails, blog posts (like this one), computer programs, software code development, how associates search for knowledge, or the creation of 3D assets (like POINT-E) for virtual worlds. Moreover, gathering feedback is essential when dealing with customers. Bosch can implement chatbots which encourage customers to fill out surveys after they chat with the support or purchase a product.

Non-technical associates can also benefit from these opportunities as only little prior knowledge is needed to use generative AI in the business context. However, users need to learn how to apply generative AI and write successful queries. For sure, humans will remain at the center of the action. They are the ones who decide, whether the AI-based concept, text, or image is valid and in line with the core company values. AI offers a significant potential to corporates such as Bosch. Our task is to make the most and the best of its opportunities

If these AI models continue to evolve (e.g., the release of increasingly larger LLMs like GTP-4), we are facing both exciting and challenging times. Just imagine all the opportunities we have before us. But let’s not forget the implications they may entail. On the one hand, these systems will undoubtedly have profound and unexpected impacts on content ownership and intellectual property protection. On the other hand, they will also stimulate a shift in knowledge and creative activity.

Whether we like it or not, generative AI is here to stay.

* ChatGPT is a free chatbot that can produce responses to practically any question. OpenAI made it available to the general public for testing in November 2022. ChatGPT is a variation of OpenAI’s GPT language model’s third generation (GTP-3), which is one of the most powerful language models available today. ChatGPT is designed specifically for chatbot applications. It must have been trained on a large dataset of conversational text.

** Due to ethical considerations, this specific example no longer works on recent releases.