Taste Labs Aims to End 'AI Slop' with New Tools

Their goal is to inject creativity and distinctiveness into AI-generated output.

"I'm Tais. I'm the founder of Taste Labs. Uh for those of you who don't know us, we came out of Stealth a few weeks ago and our whole mission is basically how do we end AI slop?"

This collaboration involves constructing appropriate post-training data or environmental adjustments to resolve identified issues. The focus is on understanding the nuances of model breakdowns to build more robust and effective AI systems.

Their approach differentiates between objective and subjective design elements. For example, alignment might be an objective design choice, while aesthetic appeal is more subjective.

Deterministic aspects, such as color palettes, contrast, and alignment, can yield objective answers if defined with sufficient specificity and context. However, subjective aspects, like aesthetics, frequently lead to disagreement even among experts.

In such cases, experts must rely on data to resolve disagreements, rather than purely subjective opinions. This methodology helps to clarify which elements of design can be objectively measured and improved by AI.

Taste Labs aims to prevent "model collapse," a phenomenon where standard AI models default to generic outputs, losing unique brand styles. This requires maintaining brand preferences and user adherence without directly altering the underlying model. The focus is on implementing context, judgment, and verification mechanisms at the application layer to guide AI output.

This approach ensures that AI-generated content remains consistent with a brand's established identity and avoids the common pitfalls of repetitive, uninspired results.

AI frequently struggles with producing content that is not merely average but genuinely unique and out of distribution. This gap highlights a core challenge for AI in generating truly exceptional creative work.

Greatness in creative endeavors involves uniqueness, meticulous attention to detail, craftsmanship, and authenticity. These qualitative aspects are difficult for AI to replicate, leading to the prevalent issue of "AI slop."

While defining "greatness" can be challenging, identifying "slop" is often straightforward because it lacks creative intentionality.

AI has democratized content creation, allowing non-experts to produce vast amounts of material. However, this ease of production has also led to a widespread prevalence of average, uninspired, and repetitive results. This abundance of generic content is what Taste Labs aims to counteract.

"I think it is hard to define what is great sometimes, but I think it's pretty pretty easy to define what is slop in the sense that most people would agree. I think the sense of like repetition of kind of soullessness is something that all of us feel right now when using AI."

Taste Labs developed a methodology involving "probes" to identify AI-generated "slop" by first performing pattern mining on extensive data to extract distinct site characteristics. This process focused on objective metrics such as colors, typography, layout, and audience, allowing for a systematic analysis of design elements.

Subsequently, they created 'baby classifiers' or probes designed to flag specific features indicative of AI slop. By combining frequency analysis, these probes achieved high accuracy in predicting the likelihood of a site being AI-generated.

The performance of this system significantly surpassed traditional "LLM as a judge" methodologies, demonstrating a more robust and reliable approach to distinguishing AI-generated content from human-created work.

Judgment is defined as the essential ability to discern, break down complex problems, and arrive at effective solutions. This highlights a shift in focus from mere production volume to qualitative decision-making.

"As the cost of production basically goes to zero. I think the thing that becomes expensive and matters more than ever is judgment."

While improving AI models is crucial, it is not sufficient to eliminate "slop" entirely. Thais Castello Branco argues that inference time, the moment of interaction with the end user, is equally if not more critical. This is where the nuanced understanding of context and intent occurs, directly impacting the quality and relevance of AI output.

Therefore, a significant portion of the effort to combat AI slop must be directed at the end-user interaction layer, ensuring that AI agents can accurately interpret and respond to user needs in real-time.

Taste Labs is developing what they've nicknamed the "Creativity API," a system designed to function as an "inspiration machine" for AI agents. The primary goal of this API is to enable AI to produce outputs that are "out of distribution," moving beyond merely average or repetitive results. This involves more sophisticated techniques than simple random temperature adjustments.

Achieving true creativity requires understanding domain-specific rules and then intentionally breaking them in strategic ways. The Creativity API aims to provide AI agents with the intelligence to make these deliberate creative choices, fostering innovation rather than uniformity.

Randomness alone can lead to outputs that feel off-brand or incoherent, failing to meet user needs.

Therefore, the strategy is to allow AI to diverge on specific elements while consciously maintaining a core set of features that align with the expected characteristics of a given category. This ensures that creative choices enhance rather than detract from the overall quality and relevance of the output.

Great branding is the culmination of deep craft and human thought, representing significant investment and strategic effort. However, companies frequently fail to translate this high-quality branding into the outputs generated by their AI agents, leading to a noticeable gap in consistency and quality. This "brand endurance" problem is identified as a major challenge.

This gap presents a significant opportunity for improvement in the quality of AI-generated content. By ensuring that AI agents can consistently reflect and maintain established brand identities, the overall standard of AI output can be significantly elevated.

The objective is to move away from generic "slop" and toward structured, high-fidelity execution.

By providing AI agents with concrete, actionable brand data, the Brand API ensures that AI-generated content is not only creative but also consistently aligned with a company's visual and thematic identity. This bridges the gap between sophisticated branding efforts and AI production.

This index offers an alternative to low-quality generative approaches, providing readily available and thoughtfully designed brand frameworks.

For example, if a user desires a "dreamy" aesthetic, they can retrieve a pre-designed dreamy brand system that has already been conceptualized to be cohesive. This approach effectively bypasses the "pillars of slop" often produced by generic AI outputs, ensuring higher quality and more intentional design outcomes.

Standard AI generation, when applied to create content, typically resulted in generic slide decks that lacked distinct branding. This illustrates a common problem where AI outputs fail to capture specific brand elements.

In contrast, an extraction-based approach, utilizing the Brand API, produced results with significantly higher fidelity to the original branding. The API accurately interpreted and applied the core elements of the General Intelligence Company's visual and stylistic identity, yielding outputs that were far more aligned and recognizable.

The current challenge in AI output is not about achieving the pinnacle of human taste, but rather about simply raising the baseline quality. Thais Castello Branco emphasizes that society has not yet earned the right to debate sophisticated questions of AI taste when basic quality standards are often unmet. The immediate goal is to elevate the current standard of AI-generated content from its foundational level.

This involves systematically decomposing complex problems and developing effective methods to measure quality. By focusing on these fundamental steps, AI outputs can move beyond mere functionality to achieve a higher, more consistent level of quality, making them more useful and reliable.