--- title: Ending AI Slop — Thais Castello Branco, Taste Labs description: Thais Castello Branco, founder of Taste Labs, presents a framework for improving AI performance in subjective domains like design and creative writing. She... canonical: https://daily.dev/posts/ending-ai-slop-thais-castello-branco-taste-labs-dlmlaqrfc twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Ending AI Slop — Thais Castello Branco, Taste Labs | daily.dev og:description: Thais Castello Branco, founder of Taste Labs, presents a framework for improving AI performance in subjective domains like design and creative writing. She... og:url: https://daily.dev/posts/ending-ai-slop-thais-castello-branco-taste-labs-dlmlaqrfc og:image: https://api.daily.dev/og/posts/dlmLAqrFc.png og:image:alt: Ending AI Slop — Thais Castello Branco, Taste Labs og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Ending AI Slop — Thais Castello Branco, Taste Labs **[AI Engineer](https://daily.dev/sources/aidotengineer)** · 16 min read · 1 upvotes · 1 comments ## Summary Thais Castello Branco, founder of Taste Labs, presents a framework for improving AI performance in subjective domains like design and creative writing. She argues that AI excels at coding and math because those domains are verifiable and measurable, while subjective domains suffer from 'collapse to the mean' — models predicting the most likely output rather than the optimal one. Her approach involves decomposing fuzzy subjective problems (e.g., brand adherence) into verifiable components that can be used as ground truth for RL environments or high-quality training data. She also emphasizes the importance of expert-curated, high-signal preference data that captures multi-preference distributions rather than averaging them into noise, advocating strongly for quality over quantity in training data for subjective domains. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### why do AI models struggle with creative writing and design compared to coding and math Code and math are verifiable, decomposable, and executable, which makes it straightforward to train and measure model performance on them. Design and creative writing lack a single correct answer, are contextual, and change over time, making it hard to define what 'good' means or to verify outputs, so models trained to predict the statistically likely output tend toward generic, mediocre results. _Anyone weighing AI tools for design or writing work can track how vendors tackle this gap on daily.dev._ ### why does AI generated content often feel like generic slop Models predict the statistically most likely next output and treat that as the ideal answer, which works for math or code but not for subjective creative work, where the most likely output is rarely the best one. Genuine creativity and greatness tend to occur at the extremes of a distribution, not the average, so optimizing for the mean produces repetitive, homogenous 'slop' rather than distinctive work. _Developers evaluating AI writing or design tools can follow this mode-collapse problem as it plays out on daily.dev._ ### how can you turn a subjective quality like brand adherence into something an AI model can be trained on Decompose the brand into concrete, codified elements such as colors, typography, motion, animation, and textures, then use that decomposition as ground truth to grade new outputs, rather than comparing them only to the original. This turns a fuzzy judgment like 'is this on brand' into something verifiable enough to shape as a reinforcement learning environment, avoiding the reward hacking and hallucination issues common with using an LLM as judge. _Teams building agents that generate on-brand assets can dig into this decomposition approach via daily.dev._ ## Community discussion Top comments from developers on daily.dev. **@agustinbarrientos** · 0 upvotes > Expert disagreements belong in the data, because averaging them away just teaches the taste model to pretend one correct answer exists. ## Similar posts on daily.dev - [Good Taste the Only Real Moat Left](https://daily.dev/posts/good-taste-the-only-real-moat-left-i2zyevivo) · Hacker News · 1 upvotes · 0 comments - [Without judgment, taste, and experience, AI simply scales mediocrity.](https://daily.dev/posts/without-judgment-taste-and-experience-ai-simply-scales-mediocrity--p3rrhqgjm) · Medium · 1 upvotes · 1 comments --- Tags: [#machine-learning](https://daily.dev/tags/machine-learning), [#llm](https://daily.dev/tags/llm), [#genai](https://daily.dev/tags/genai), [#reinforcement-learning](https://daily.dev/tags/reinforcement-learning) [View this post on daily.dev](https://daily.dev/posts/ending-ai-slop-thais-castello-branco-taste-labs-dlmlaqrfc) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Ending AI Slop — Thais Castello Branco, Taste Labs","url":"https://daily.dev/posts/ending-ai-slop-thais-castello-branco-taste-labs-dlmlaqrfc","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/ending-ai-slop-thais-castello-branco-taste-labs-dlmlaqrfc"},"datePublished":"2026-07-31T19:31:22.206Z","dateModified":"2026-09-14T08:34:34.397Z","description":"Thais Castello Branco, founder of Taste Labs, presents a framework for improving AI performance in subjective domains like design and creative writing. She...","image":"https://i.ytimg.com/vi/lCBf9slCanI/sddefault.jpg","thumbnailUrl":"https://i.ytimg.com/vi/lCBf9slCanI/sddefault.jpg","isAccessibleForFree":true,"articleSection":"AI Engineer","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"AI Engineer","logo":"https://media.daily.dev/image/upload/s--u5PucxNT--/f_auto/v1724338940/logos/aidotengineer","url":"https://daily.dev/sources/aidotengineer"},"commentCount":1,"discussionUrl":"https://daily.dev/posts/ending-ai-slop-thais-castello-branco-taste-labs-dlmlaqrfc","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":1}],"keywords":"machine-learning,llm,genai,reinforcement-learning","timeRequired":"PT16M","video":{"@type":"VideoObject","name":"Ending AI Slop — Thais Castello Branco, Taste Labs","description":"Thais Castello Branco, founder of Taste Labs, presents a framework for improving AI performance in subjective domains like design and creative writing. She...","thumbnailUrl":"https://i.ytimg.com/vi/lCBf9slCanI/sddefault.jpg","uploadDate":"2026-07-31T19:31:22.206Z","duration":"PT16M","url":"https://api.daily.dev/r/dlmLAqrFc","embedUrl":"https://www.youtube.com/embed/lCBf9slCanI"}} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"AI Engineer","item":"https://daily.dev/sources/aidotengineer"},{"@type":"ListItem","position":3,"name":"Ending AI Slop — Thais Castello Branco, Taste Labs"}]} {"@context":"https://schema.org","@type":"WebPage","@id":"https://daily.dev/posts/ending-ai-slop-thais-castello-branco-taste-labs-dlmlaqrfc","comment":[{"@type":"Comment","text":"Expert disagreements belong in the data, because averaging them away just teaches the taste model to pretend one correct answer exists.","datePublished":"2026-08-03T02:33:44.781Z","url":"https://daily.dev/posts/dlmLAqrFc#c-TFqHeLLJ9","author":{"@type":"Person","name":"Agustin Barrientos","url":"https://daily.dev/agustinbarrientos","image":"https://media.daily.dev/image/upload/s--5ayxQnqn--/f_auto/v1788281802/avatars/avatar_wQYYVe5Tbj0NJ7C7qPoa8?_a=BAMAMicg0"}}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/ending-ai-slop-thais-castello-branco-taste-labs-dlmlaqrfc#faq","mainEntity":[{"@type":"Question","name":"why do AI models struggle with creative writing and design compared to coding and math","acceptedAnswer":{"@type":"Answer","text":"Code and math are verifiable, decomposable, and executable, which makes it straightforward to train and measure model performance on them. Design and creative writing lack a single correct answer, are contextual, and change over time, making it hard to define what 'good' means or to verify outputs, so models trained to predict the statistically likely output tend toward generic, mediocre results. Anyone weighing AI tools for design or writing work can track how vendors tackle this gap on daily.dev."}},{"@type":"Question","name":"why does AI generated content often feel like generic slop","acceptedAnswer":{"@type":"Answer","text":"Models predict the statistically most likely next output and treat that as the ideal answer, which works for math or code but not for subjective creative work, where the most likely output is rarely the best one. Genuine creativity and greatness tend to occur at the extremes of a distribution, not the average, so optimizing for the mean produces repetitive, homogenous 'slop' rather than distinctive work. Developers evaluating AI writing or design tools can follow this mode-collapse problem as it plays out on daily.dev."}},{"@type":"Question","name":"how can you turn a subjective quality like brand adherence into something an AI model can be trained on","acceptedAnswer":{"@type":"Answer","text":"Decompose the brand into concrete, codified elements such as colors, typography, motion, animation, and textures, then use that decomposition as ground truth to grade new outputs, rather than comparing them only to the original. This turns a fuzzy judgment like 'is this on brand' into something verifiable enough to shape as a reinforcement learning environment, avoiding the reward hacking and hallucination issues common with using an LLM as judge. Teams building agents that generate on-brand assets can dig into this decomposition approach via daily.dev."}}]} ```