I might be one of those typical skeptical Germans, often associated with the characteristic German angst. The year 2022 concluded with the groundbreaking introduction of ChatGPT. While I was familiar with models like Stable Diffusion and others before, ChatGPT marked a significant milestone. Initially, it invoked a sense of fear and uncertainty in me as a CEO, as it often does when faced with the unknown and the possibility of being replaced, which is a distinctly human reaction.

The fear stems not only from the uncertainties associated with the introduction of groundbreaking technologies but also from the broader implications it may have on the company’s stability. The prospect of being replaced by advanced AI models can be particularly daunting, raising questions about the company’s resilience and the adaptability of its existing structures.

The year 2023 proved to be a challenging journey, and from a psychological standpoint, I was experiencing all the five stages of coping.

1. “AI is just a new Trend!”| Denial

In the beginning, I was skeptical about the profound impact of artificial intelligence in its current stage. I used to reassure myself with the belief that it was merely a matter of statistics and nothing more. At the time, I couldn’t foresee the transformative power that lay within the evolving realm of AI. Little did I know that beyond the statistical foundations, AI would unfold into a revolutionary force, reshaping industries and redefining the way we interact with technology. My initial skepticism gave way to a realization of the vast potential that AI holds in our rapidly changing landscape.

2. “We missed an incredible opportunity.”| Anger

My primary source of frustration stems from the fact that I delved deeply into developing perceptron networks around 2001, a time when I was nominated for a student scholarship by Professor Dr. Otte for the ‘Stiftung des Deutschen Volkes.’ During that period, I was actively involved in creating image generation networks and constructing neural networks from the ground up. The question that nags at me is, why didn’t I push myself and our company to continue with that wealth of knowledge?

3. “We’re not actually that bad…”| Bargaining

At that stage, I bargained with myself. It wasn’t that bad. After all, we’ve been at the forefront, specializing in companion software for AI for years. Nearly every app and desktop application we’ve developed over the last five years has, in some way, been intertwined with artificial intelligence. This has encompassed an array of cool functionalities, from feature extraction in audio to image segmentation, image feature extraction, 3D reconstruction from 2D images, and even combining augmented reality with AI, among other exciting ventures

4. “We are making ourselves dependent!”  | Depression

Yes, we’ve utilized numerous top-notch AI technologies, but often in a hands-off manner. We encountered a challenge that’s prevalent among many European companies. Take Siemens, for instance, which recently announced a partnership with AWS (see: https://press.siemens.com/global/en/pressrelease/siemens-and-aws-join-forces-democratize-generative-ai), increasing their dependency on an external driver.

Similar to many, we found ourselves reliant on major players like OpenAI. Our work was overly dependent on the contributions of others, leaving us without the autonomy to propel it forward.

5. “Let’s build AI-Models ourselves!” | Acceptance

Finally, in March 2023, we accepted the current state and decided to delve deeper into the game using our own means. While we couldn’t reinvent the wheel, we discovered the value of leveraging existing designs that we could adapt for our applications. Since then, we’ve been internally driving LLM training, like LLAMA, and training StableDiffusion Models. Although LLAMA wasn’t production-ready at that time, we were eager to get our hands dirty and dive into the code.

From then on, we’ve employed Dolly, StableLM, Bloom, Vicuna, and LLAMA2 with our own datasets to build and fine-tune exceptional LLMs, especially tailored for embedding in apps or offline hardware. This approach enables us to achieve greater speed (in terms of responsiveness) and security (as no external access is needed). Importantly, it also ensures data confidentiality—you won’t have to disclose your training data to external parties

Final Conclusion

Today, I am very certain that every software company needs to adapt to the new environment. Utilizing AI in various fields is a great opportunity to enhance your development speed since you don’t need to worry about complex mathematics, such as in beat detection in audio. Instead, you focus on the data and the training of those models. It is a powerful tool to expedite software development and ensure reliability. Nevertheless, it is crucial to understand the limitations and to assist customers in decision-making. Not every brilliantly usable application needs AI, but any AI application requires a great user experience and an application that effectively leverages it.

To answer the question, do we need to fear AI? Of course not. We should embrace it and use it as the tool it is.

Häufig gestellte Fragen: