One technological innovation that really stood out last year was the release and widespread accessibility of a type of artificial intelligence called ChatGPT, published by the company OpenAI. And although it had been apparent for some time in the scientific field that a major breakthrough that could be described as revolutionary was imminent, it was surprising and overwhelming for the rest of the world to have a "thing" available that could be described as intelligent in the true sense of the word. The impact of this event can be seen from the fact that after OpenAI made the software version GPT-3 available to the public free of charge on November 30, 2022, one million users worldwide signed up within five days.

Technically speaking, ChatGPT can be described as a Large Language Model (LLM). LLMs are part of Natural Language Processing (NLP) research, which is concerned with the development of models that can understand and generate natural language. LLMs have been around for several years, but only with the development of very large models such as GPT-3 and ChatGPT with 175 billion parameters have they really gained practical importance. These models can solve multi-level complex tasks - from answering questions to generating texts - without any special preparation or model training via textual queries, so-called prompts.

But what also quickly became clear: Despite their power, there are still many challenges that need to be overcome in order to use ChatGPT sensibly and efficiently. On the one hand, this means that we as people and users have to learn how to interact with this technology. We need to learn how to assess the results and find the best prompt in each case. And last but not least, we need to find answers to the question of how best to integrate these models into products, company processes and business models.

Prompt engineering as an elementary tool

Exemplary information flow from prompt creation to transfer to a Large Language Model, which then generates the final output. Sequence of prompt engineering and interaction with LLMs.

© Source: talsen team

Providing an overview of possible areas of application for this technology in industrial companies is not that easy. This is because requests or prompts can be made in almost any kind of natural language on practically any subject area. This offers a new and enormous potential to use this capability company-wide to develop products and services, convert documents and technologies, extract information of all kinds, improve customer service or optimize marketing measures. There are definitely no limits to the creativity for meaningful use cases here!

Alongside generative text creation, similar tools are also being developed for the automated recognition and creation of images, videos and sound. We can therefore assume that we will be living in a generative and multimodal world in the future and that we will be able to draw on this potential. The paradox of using generative AI may be that, despite the supposedly simple access, special technical skills are still required in order to achieve a specific result that meets one's own expectations with the textual descriptions of a prompt. The need for special skills has even led to a new job description - prompt engineering - and a corresponding job title - prompt engineer. However, this refers less to the one-off creation of a simple query in a web browser and more to the creation of reliable, reusable and robust prompts that can be used in a productive environment.

How is generative AI changing the company?

The chronological development of LLMs up to the present day based on important preliminary developments and a selection of models.

© Source: promptengineering.org

However, the challenge does not end with simply knowing how to formulate a good prompt. The now very easy access to information and skills that were previously reserved for certain roles such as marketing, sales, product management or development are now very simply available to more or less "everyone". This will inevitably lead to a change in roles and responsibilities; above all, the processes that are already networked today will become even more interconnected and overlap. The result will ultimately lead to a further increase in corporate dynamics. However, to avoid building castles in the air in the truest sense of the word, it is important to know not only the results of the respective prompts, but also the prompt itself and its history.

This gives the prompt the same, if not greater, importance than the answer itself. This is because the answer can be regenerated from a good prompt at any time. This makes prompt change management one of the biggest challenges. It is therefore advisable to treat every prompt like software code and to version it.

What can we expect from LLMs in 2024?

The author: Dr.-Ing. Hans Egermeier is Managing Director of talsen team.

© talsen team

This year, we will see a rapidly accelerating development in the world of artificial intelligence. Especially in terms of merging the different modalities of text, image, video and audio. It can be assumed that we will quickly become accustomed to the chatbots that are now available everywhere and that they will increasingly be seamlessly integrated into all the productive tools we are familiar with. As already started by OpenAI, this year will also see the emergence of store concepts based around intelligent applications and chatbots. This will also have a major impact on new and different business models, as the commercially usable models incur fees per prompt/response. It can also be assumed that the systematic and standardized evaluation of the quality and usefulness of chatbots and generative AI tools will become increasingly important. Experimental trial and error prompt engineering - which many people probably experienced last year with the web interface and GPT 3.5 - will gradually give way to targeted optimization approaches. Last but not least, it is already foreseeable that the combination of several generative AIs with different roles for generating, evaluating and optimizing results and solution strategies will be the next big step. These emerging agents will significantly increase the complexity and depth of solutions and automate further processes that can currently only be achieved with human interaction.