Ending AI Slop — Thais Castello Branco, Taste Labs
Thais Castello Branco from Taste Labs argues that AI models today are poor at subjective work such as writing and design, and that the problem is that models tend to optimize toward the average — the most likely outcome — which kills creativity.
The solution is to break down design into measurable components (a logo, a headline, a color) that can be compared against the original, and then gather high-quality feedback from experts linked to specific choices. By making taste measurable and trainable in this way, AI can learn to break from the obvious when the situation requires it.
Open on YouTube →
Vibekollen prepared this summary with AI from the original publication. The content belongs to AI Engineer.
