Specialized AI for design critique

The idea was simple yet audacious: Harness the analytical power of generative AI to provide the kind of feedback designers crave — honest, constructive, and, most importantly, actionable. No more vague compliments. No more beating around the bush. Just straight-up, valuable insights to elevate our designs.

All of this, packaged in a specialized version of ChatGPT, designed to do what humans struggle with: providing detailed, unbiased, and brutally honest UX feedback.

The challenge was to test if the visual capabilities of GPT4 were already good enough to understand the nuances of UX design. And let me tell you, the results were mind-blowing.

Defining good UX feedback

First, we needed to fine-tune the GPT model with what actually makes good feedback. It’s not just a matter of spitting out data; it’s about understanding the art of critique.

We started with the rose, thorn, bud technique as a framework for feedback. The rose highlights what’s working, the strengths that give life to your design. The thorn, that’s the critical part, pinpointing exactly what needs fixing, no sugarcoating. And the…