Killing the 'Slop': How Taste Labs Uses AI to Save Web…

Explore how Taste Labs fights the homogenization of the internet using 'probes' and structured brand components to eliminate AI-generated slop.

Ever feel like every single website you visit looks exactly the same lately? It's not just your imagination; it's a phenomenon called 'slop,' and while AI didn't start the fire, it's definitely pouring gasoline on it.

⚡ TL;DR: Taste Labs is fighting the 'homogenization' of the web. Instead of using one big AI to judge design, they use a fleet of small classifiers called 'probes' to detect and eliminate generic, context-blind AI design.

The Simple Version

For years, the internet has been becoming boring. Palettes and layouts started converging long before ChatGPT, but AI accelerated this by removing context.

AI doesn't naturally understand that a high-end law firm should feel different from a local pet shop. It just knows what a 'generic website' looks like, resulting in a sea of sameness that Thais Castello Branco calls 'slop.'

🎯 Think of it this way: Imagine an AI chef who has memorized every recipe in the world but has no taste buds. It can make a cake and a steak, but it might put frosting on the steak because 'both are food'—it lacks the judgment to know why they shouldn't mix.

How It Actually Works

Defining 'Slop'

To fix a problem, you first have to measure it. Taste Labs identifies three specific signatures of slop: repetition, lack of fit, and low intent.

While 'taste' feels subjective, much of design is actually deterministic. Things like contrast, alignment, and color theory are objective rules that most experts agree on.

🎯 Think of it this way: It's like a building inspection. An inspector doesn't ask if a house is 'pretty' (subjective); they check if the wiring is safe and the foundation is level (objective).

The Power of Probes

Instead of asking a large language model (LLM) to 'judge' if a site looks good, Taste Labs uses probes.

Probes are small, specialized classifiers. Each one is trained to detect just one specific characteristic of a design.

By stacking these probes together, the system can predict 'slop' with much higher accuracy than a general-purpose AI model could.

Technical Deep Dive

From a developer's perspective, the magic happens in how Taste Labs handles inference and distribution. Rather than simply raising the 'temperature' of a model to get more creative results—which often just leads to random chaos—they push the model out of distribution on purpose.

This means the system is designed to break specific rules while still respecting the overarching expectations of a category. It's a controlled rebellion against the mean.

Furthermore, they are moving away from 'prompting' and toward structured brand components. By turning a brand's identity into a set of structured data, an AI agent can follow a strict rulebook and be graded against it objectively.

Why This Matters

If we rely solely on generative AI for design, we risk a 'collapse' of visual diversity. When every brand uses the same AI-generated templates, they lose their unique identity and the ability to build trust through distinct aesthetics.

By making 'taste' measurable through data and probes, we can move toward AI that doesn't just generate content, but understands intent and context.

Key Takeaways

Homogenization: The web was becoming generic before AI, but AI made it context-blind.
Slop Signatures: Defined by repetition, lack of fit, and low intent.
Probes vs. LLMs: Small, specialized classifiers beat general models at judging design quality.
Objective Design: Contrast and alignment are deterministic; aesthetics are data-driven.
Controlled Creativity: Pushing 'out of distribution' is better than simply raising model temperature.
Structured Branding: Turning brands into components allows agents to be graded for accuracy.

Killing the 'Slop': How Taste Labs Uses AI to Save Web Design