Why AI Always Creates The SAME Boring Designs (And How to Fix It)

Understanding AI Output and Prompting Techniques

The Problem with AI Outputs

• AI models often produce similar outputs characterized by common design elements like purple gradients and generic fonts due to their training on widely available internet data.

• The training process can be visualized as a funnel where the AI absorbs vast amounts of data, leading to repetitive patterns in its outputs.

• When prompted vaguely, the AI tends to gravitate towards average responses, which may include overused phrases or styles that it encountered frequently during training.

Introducing Prompting Altitude

• The concept of "prompting altitude" refers to finding a balance between vague and overly specific prompts when interacting with AI.

• High altitude prompts are too vague (e.g., simply asking for a report rewrite), while low altitude prompts are excessively rigid (e.g., specifying every detail).

Finding the Goldilocks Zone

• The goal is to achieve a middle ground—neither too high nor too low—in order to elicit unique and relevant outputs from the AI.

• Effective prompting involves providing both direction (what to avoid or pursue in design choices) and inspiration (general ideas that guide creativity).

Practical Application of Prompting Techniques

• Direction should focus on avoiding common pitfalls, such as typical color schemes or layouts seen in many AI-generated designs.

• Inspiration can be drawn from well-known brands or aesthetics without needing extensive design knowledge; simple references can guide the output effectively.

Utilizing Cloud Skills for Enhanced Outputs

• A "cloud skill" is defined as a collection of prompts organized into folders that an AI can access, enhancing its ability to generate tailored designs.

• These skills allow users to create unique outputs by steering clear of generic results typically produced by standard prompts.

Creating Impressive Designs with AI Skills

Introduction to AI Skills

• The speaker discusses the creation of a cloud skill aimed at enhancing design capabilities, referencing a blog post from Anthropic for guidance.

• A specific prompt from the blog post is highlighted as crucial for understanding the context and requirements for developing high-quality skills.

Skill Creation Process

• After implementing the prompt, the AI successfully initialized a skill called "creator," which generates other skills and produced two files: one detailing the skill's contents in markdown format and another containing the skill itself.

• The speaker emphasizes that these skills can be utilized across various models (e.g., ChatGPT, Gemini), as they consist of prompts bundled into files.

Managing Skills in Claude

• Users can access their created skills through settings under their profile, where both out-of-the-box and custom skills are listed.

• The speaker showcases a custom front-end design skill created to improve visual outputs for dashboards and graphics.

Understanding Skill Functionality

• The importance of applying these skills universally is reiterated, with an intention to demonstrate how they function within ChatGPT.

• Key components of the skill include prompts directing unique font choices, color depth, animations, and avoiding common design pitfalls.

Design Guidelines Within Skills

• Anti-pattern guidelines are included in the skill to prevent overused fonts (like Arial or Roboto), purple gradients on white backgrounds, and cookie-cutter layouts in dashboard designs.

Examples of Improved Outputs

• The speaker presents examples comparing outputs generated without any skills versus those enhanced by using them.

• For instance, a basic landing page created without a skill features typical AI traits like white backgrounds and purple gradients.

• When using the enabled skill with detailed prompts, it results in more visually appealing designs with unique colors and fonts.

Further Design Examples

• Another example involves creating a blog site; initially vague prompts yield generic results. However, enabling the design skill leads to more distinctive layouts that avoid common pitfalls like bland backgrounds or standard structures.

AI Dashboard Design: Enhancements and Comparisons

Overview of AI-Generated Dashboards

• The speaker discusses the typical appearance of AI-generated dashboards, highlighting common features such as purple gradients and basic fonts.

• A comparison is made between a generic dashboard layout and one enhanced by specific skills, noting that the latter offers more unique colors and attractive buttons.

Example of Enhanced Output

• The speaker shares an example where a prompt from Sam Altman was used to create a chart based on energy usage over time.

• An initial output without enabling any skills is shown, characterized by a white background and standard design elements typical of AI outputs.

Utilizing Skills for Improved Design

• By adding an additional sentence to the prompt instructing the use of a newly created skill, the output's design improves significantly.

• The enhanced version includes better font choices, intuitive chart designs with labeled timelines, and animations that engage users.

Comparison with Chatbot Outputs

• The same blog post was used in a chatbot context to generate another chart; initial results showed some improvements but also highlighted issues like text overlay on lines.

• In this case, while there were advantages such as no white background, certain aspects still indicated an AI origin.

Final Enhancements and Key Takeaways

• After uploading the Claude skill into the chatbot environment and refining prompts further, significant improvements were noted in clarity and intuitiveness of charts.

• The speaker emphasizes the importance of crafting prompts at varying altitudes—balancing specificity with flexibility—to achieve optimal results across different AI applications.