Taste Labs invests in specialized data to teach AI how to make subjective evaluations

Brazilian startup Taste Labs has raised $18.5 million in venture capital to develop a new approach to artificial intelligence: teaching machines to make subjective decisions, such as evaluating aesthetics, design, and creativity. The company believes that the next breakthrough in AI will not depend solely on larger models, but on specialized data produced by human experts who can distinguish generic creations from those of higher quality.

The problem of “AI Slop” and the quest for quality

In the age of artificial intelligence, users have started looking for clues in images, videos, and texts to determine whether content was generated by machines. The phenomenon known as “AI Slop”—low-quality, mass-produced content—has spread across the internet. With its $18.5 million in funding, Taste Labs aims to combat this trend by improving AI’s ability to produce results that are clearly good.

“We want to kill AI slop,” says Marcelo Afonseca, a founding member of Taste Labs. “It represents a convergence to the average, where everything looks the same, and AI introduces elements that make it obvious that they were generated by machines.”

Teaching taste to a machine: the role of tastemakers

According to the company, the main bottleneck of current AI is not just the increase in computational power or the size of models, but the difficulty of evaluating quality in areas without an objective answer. In fields such as design, a technically functional solution may still look generic, while a well-crafted presentation can convey visual identity and purpose.

To overcome this limitation, Taste Labs created a network of designers, called tastemakers, who evaluate AI model outputs and transform human criteria—such as aesthetics, composition, and visual coherence—into structured data. This data is used to train and refine AI systems, teaching them to recognize and reproduce patterns of quality.

Partnerships with laboratories and the importance of specialized data

The startup collaborates with AI laboratories to conduct evaluations, benchmarks, and post-training adjustments. Afonseca explains that after the initial phase of generative models like ChatGPT, researchers’ trust became essential. “We built relationships with the labs, bringing our opinion and specific knowledge,” he says.

The work involves comparing models in different formats—websites, presentations, and applications—using designers’ evaluations to measure quality. In these comparisons, the difference between current performance and the level expected by professionals becomes evident.

Future vision: data as a strategic layer of AI

In Afonca’s view, the evolution of AI should increase the value of specialized data. Competition will not be just about building larger models, but about the ability to provide information that improves performance in specific tasks. He compares the technology to a chain formed by layers: model developers, infrastructure providers, and companies capable of creating proprietary data.

“The data layer will become increasingly important, and whoever knows how to develop it and use it well will be more relevant,” he says.

The human as evaluator and guide of the machine

Despite the advancement of generative models, Taste Labs believes that human experts will remain essential in areas without a single answer. “We will always need humans to help, especially with subjective things. There are many designs created by AI, and the vast majority are bad. Whoever can distinguish the good from the bad is becoming increasingly important,” says Afonseca.

The role of creative professionals may shift from creator to evaluator and guide of systems. The startup plans to expand its approach to music, film, and creative writing, where quality evaluation also depends on subjective criteria.

Frequently Asked Questions

• What is Taste Labs? A Brazilian AI startup that has raised $18.5 million and works with specialized data to improve the subjective judgment of AI models in design, aesthetics, music, film, and creative writing.
• What does AI Slop mean? Low-quality content produced in bulk by AI, which follows repetitive and generic patterns.
• What problem is Taste Labs trying to solve? The difficulty of current AI in evaluating quality in tasks without an objective answer, such as design and artistic creation.
• Why does AI struggle with subjective tasks? Models have evolved in areas with clear criteria, such as mathematics, whereas in creative domains there is no single answer.
• How does Taste Labs teach AI to have “good taste”? By using human tastemakers who evaluate outputs and transform subjective criteria into data for training.
• What will be AI’s next challenge? Improving the ability to make subjective decisions, with a focus on specialized data beyond larger models.
• Will humans be replaced by AI in creative fields? The company argues that professionals will remain important as evaluators and guides of systems.
• In which areas does Taste Labs plan to expand its technology? Music, film, and creative writing, in addition to design.