Content Strategy for AI: Three Things I got Wrong

🌈 Abstract

The article discusses the author's experience and insights in designing custom generative AI solutions, specifically custom GPTs on OpenAI's platform. It covers the key mistakes the author made and the lessons learned, including:

• Understanding the difference between "training" a custom GPT versus designing the knowledge base and user interface
• The importance of structuring content into a tidy database for the bot's retrieval and generation
• The need to manage multiple bots with different knowledge sources and audiences
• Challenges in organizing and tracking the connections between knowledge sources and bots

The article also touches on the broader implications of AI for UX designers, including the need to relinquish control over pixel-perfect designs and embrace personalized user experiences.

🙋 Q&A

[01] Understanding Custom GPTs

1. What is the difference between "training" a custom GPT and designing the knowledge base and user interface? The author initially thought that creating a custom GPT was about building a "mini language model" on top of the large language model. However, they later realized that the focus should be on:

• Designing a curated, structured knowledge base for the bot to retrieve information from (rather than just dumping large amounts of unstructured content)
• Crafting the user interface and prompt engineering to enable effective user interaction with the bot

2. Why is structuring content into a tidy database important for a bot's retrieval and generation? Unlike the large, messy datasets used to train LLMs, a bot's knowledge base needs to be succinct, tidy, and non-duplicative. This structured approach allows the bot to effectively "retrieve" quality information for its "augmented generation" process.

3. What are the key challenges in managing multiple bots with different knowledge sources and audiences? The author had to create several bots (e.g., a free Q&A bot, a paid OOUX assistant, an internal virtual assistant, an OOUX Masterclass librarian) with different knowledge sources and audiences. Tracking the status, maintenance plans, and connections between knowledge sources and bots became a complex challenge that required a more structured, OOUX-based approach.

[02] Implications for UX Designers

1. How does the rise of AI require UX designers to relinquish control, and what new powers does it provide? As design has evolved from print to the internet to responsive design to design systems, UX designers have had to relinquish more control over the final user experience. With AI, designers will need to relinquish even more control, as user interactions will be highly personalized and unpredictable. However, this also provides new powers, allowing designers to focus on higher-order problem-solving rather than repetitive interface design.