Had another fantastic meet-up with our partners at DottingAi. Charles, Richard, and I delved deep into exploring the boundaries of our AI bots. It's exciting to see how far we've come with automation, data warehousing, and AI integration. But now, we're facing an intriguing new challenge: defining the style in which we train our AI.

How to use visuals to make new deliverable that work? We've been pondering over the visual representation of concepts like moonshots - those 10X ideas for the future. It's fascinating how AI tends to depict the future in blue tones. But should we aim for hyper-realistic, almost photographic representations? Or would that diminish the visionary aspect, the fantasy, the ability to imagine? Perhaps a more conceptual sketch style, like those used in advertising, might be more appropriate for tools like PIVOT.

Putting your identity to words (and back again) Jeroen's meticulous attention to design details - from fonts to color, line thickness, and proportions - is really paying off now. It's providing us with a solid foundation for prompting and training the AI effectively. As Charles aptly put it, many AI algorithms excel at creativity and generation but struggle with logic and following precise instructions. One of the questions I asked Jeroen;"How would you brief a designer over the phone to create something in our style?" Design system & storyworlds More and more I'm convinced that we don't need to design logo's or ideas we need to build identities that are more like a storyworld, that are based on a design system. I will elaborate on that later, sorry!

Training a visual AI We're still working on finding a consistent approach, especially with visuals. Text is somewhat easier to manage. For now, we're feeding the algorithm with our own images. I've already uploaded more than 1000 elements from Jefta's drawings. It'll be interesting to see what the system comes up with in the coming week. The goal isn't to replace Jefta's unique vision and soul, but to create images that align closely with his style, making it easier for users to develop plans and concepts.

Meta-prompting We're also exploring meta-prompting. Crafting effective prompts is challenging - my attempts with Leonardo or Dalle-2 often result in childish drawings, while Charles' prompts are much more on point. He explains this by saying that computers are better understood by computers (my loose translation). Using meta-prompts to let AI write its own prompts is proving effective. We're also using negative prompts to specify what the AI should avoid, like text in images, which helps with use in different regions.

Why do we do this like that? I'm curious about potential adjustments for regional partners. If we expand with our partner in India, should we generate different images than in the USA or EU? It's fascinating how technical possibilities constantly drive us to question our core purpose. Why? Why? Why do we do this? Each advance brings new questions. Long live the frontier of new possibilities!

This journey is pushing us to really define our identity, both in words and visuals.

End-result for today; 100% AI.