Generative AI

I want to end the year with an article on generative AI (genAI). Although the algorithms and methods are already decades old, 2023 is when the impressive possibilities have reached the masses.

Spreadsheets and calculators have revolutionized the work of accounting and finance departments. AI will fundamentally change knowledge work. The possibilities offered by artificial intelligence are enriching and liberating. In the field of software development, AI assistants (Github Copilot, JetBrains AI Assistant, etc.) will free us from tedious routine tasks and help us to write better software (technical debt) and secure code (security vulnerabilities).

In software development, knowledge of the business domain, the possibility of software architectures, soft skills, and business analysis methodologies will become even more critical.

To show the possibilities and limitations of AI tools, I asked Google Bard, Microsoft Copilot, and OpenAI ChatGPT to create a blog about Oswald Ungers. Oswald Ungers was an architect from my hometown, Kaisersesch. He is known as a hidden champion in both the USA and Germany. He is unknown in Kaisersesch itself.

The first test’s impressive results clearly showed that Kaisersesch is not widely appearing in public sources. It led to significant detail errors in the generated blogs.

Please decide for yourself!

Sections

The test consists of five parts. Please consider that Google Bard is in the experimental phase, Microsoft Copilot is at the beginning, and it uses the free version of ChatGPT 3.5.

Part I contains three blogs created with the AI mentioned above assistants. Each AI assistant answered the same question:

Please write a blog about Oswald Ungers and the influences from his hometown. Consider his influences on modern architecture and his professorships.

The first test showed that all AI models needed in-depth knowledge of Kaisersesch. As a result of this, I asked:

Please generate a blog about the history of Kaisersesch and its architecture.

Part II ChatGPT 3.5, Part II Google Bard, Part IV Microsoft Copilot show the generated blog post.

Part V gives some hints and comments about the wrong interpretation of the models. I tried not to value the outcome.