--- schema: 1 type: video title: 'Training Taste — Thais Castello Branco, Taste Labs' description: >- Taste Labs analyzed more than two million websites going back a decade, and the finding that complicates the usual story is that the internet was homogenizing before AI arrived: palettes converging, layouts converging, trends spreading faster than they once did. AI did not start the collapse. It accelerated it and made it context blind, so a pet shop and a finance firm end up with the same page. Thais Castello Branco founded Taste Labs, out of stealth weeks before this talk, to attack that. She names three signatures of slop: repetition, lack of fit, and low intent. She is careful about the word taste, preferring judgment, and she rejects the idea that the fix is everyone acquiring some. Designers spend careers on it, through exposure, pattern recognition, restraint, and the nerve to break a rule on purpose. So the work is making a fuzzy domain measurable. Some pieces go nearly deterministic once the context is tight: contrast, alignment, and palette selection are things most experts score the same way. Aesthetics is where experts disagree, so that leans on data instead. Her team mined features from those sites and trained what she calls probes, small classifiers that each detect one characteristic. Stacked together, their combined frequency predicts slop, and she reports it beating the usual approach of asking a model to judge. On the solution side she argues inference time matters as much as the model, because intent and context are exchanged there. One product pushes deliberately out of distribution, breaking chosen rules while respecting a category's expectations rather than raising temperature. Another turns a brand into structured components an agent can follow and that you can grade it against. Speaker info: - https://x.com/thaiscbranco_ - https://www.linkedin.com/in/thais-castello-branco/ Timestamps: 0:00 - Taste Labs, and slop as the enemy 1:33 - Where design gets objective and where it does not 2:53 - Greatness, and why slop is easier to define 3:46 - Generation costs nothing, taste takes a lifetime 5:07 - Three signatures of slop 6:28 - Two million websites, and what they showed 7:48 - Probes, small classifiers that predict slop 10:31 - Going out of distribution on purpose 11:51 - Turning a brand into something an agent can follow --- Watch along with the corrected transcript, summary, timestamps and resources on this talk's official AIE page: https://ai.engineer/talks/sDMGWK4wZ_w-youtube-ec92d771f77b460b07f4 Join our upcoming AI Engineer conferences (see https://ai.engineer landing page): - AIEi Shanghai, November 5–6 — the first for meeting top Chinese labs and engineers - AIE CODE in SF, November 10–12 — the top AI Coding conference returns - AIEi Sydney, December 7–8 — the first AIEi run alongside NeurIPS 2026 in Sydney --- canonical: >- https://liquidgold.cc/channel/aidotengineer/explore/video/training-taste-thais-castello-branco-taste-labs-c149503859e1 markdown: >- https://liquidgold.cc/channel/aidotengineer/video/training-taste-thais-castello-branco-taste-labs-c149503859e1.md llms: null parent: 'https://liquidgold.cc/channel/aidotengineer.md' channel: aidotengineer updated: '2026-10-24T03:48:43.296Z' language: null children: [] videoId: sDMGWK4wZ_w durationSeconds: 906 publishedAt: '2026-09-10T16:00:20.000Z' searchChunks: - ordinal: 0 text: null startSeconds: 22 endSeconds: 100 watch: 'https://liquidgold.cc/watch/sDMGWK4wZ_w?t=22' heldReason: rights - ordinal: 1 text: null startSeconds: 22 endSeconds: 100 watch: 'https://liquidgold.cc/watch/sDMGWK4wZ_w?t=22' heldReason: rights - ordinal: 2 text: null startSeconds: 22 endSeconds: 100 watch: 'https://liquidgold.cc/watch/sDMGWK4wZ_w?t=22' heldReason: rights - ordinal: 3 text: null startSeconds: 111 endSeconds: 178 watch: 'https://liquidgold.cc/watch/sDMGWK4wZ_w?t=111' heldReason: rights - ordinal: 4 text: null startSeconds: 191 endSeconds: 233 watch: 'https://liquidgold.cc/watch/sDMGWK4wZ_w?t=191' heldReason: rights - ordinal: 5 text: null startSeconds: 244 endSeconds: 307 watch: 'https://liquidgold.cc/watch/sDMGWK4wZ_w?t=244' heldReason: rights - ordinal: 6 text: null startSeconds: 319 endSeconds: 396 watch: 'https://liquidgold.cc/watch/sDMGWK4wZ_w?t=319' heldReason: rights - ordinal: 7 text: null startSeconds: 408 endSeconds: 476 watch: 'https://liquidgold.cc/watch/sDMGWK4wZ_w?t=408' heldReason: rights - ordinal: 8 text: null startSeconds: 476 endSeconds: 487 watch: 'https://liquidgold.cc/watch/sDMGWK4wZ_w?t=476' heldReason: rights - ordinal: 9 text: null startSeconds: 487 endSeconds: 639 watch: 'https://liquidgold.cc/watch/sDMGWK4wZ_w?t=487' heldReason: rights - ordinal: 10 text: null startSeconds: 487 endSeconds: 639 watch: 'https://liquidgold.cc/watch/sDMGWK4wZ_w?t=487' heldReason: rights - ordinal: 11 text: null startSeconds: 487 endSeconds: 639 watch: 'https://liquidgold.cc/watch/sDMGWK4wZ_w?t=487' heldReason: rights - ordinal: 12 text: null startSeconds: 639 endSeconds: 652 watch: 'https://liquidgold.cc/watch/sDMGWK4wZ_w?t=639' heldReason: rights - ordinal: 13 text: null startSeconds: 652 endSeconds: 717 watch: 'https://liquidgold.cc/watch/sDMGWK4wZ_w?t=652' heldReason: rights - ordinal: 14 text: null startSeconds: 728 endSeconds: 906 watch: 'https://liquidgold.cc/watch/sDMGWK4wZ_w?t=728' heldReason: rights - ordinal: 15 text: null startSeconds: 728 endSeconds: 906 watch: 'https://liquidgold.cc/watch/sDMGWK4wZ_w?t=728' heldReason: rights - ordinal: 16 text: null startSeconds: 728 endSeconds: 906 watch: 'https://liquidgold.cc/watch/sDMGWK4wZ_w?t=728' heldReason: rights --- # Training Taste — Thais Castello Branco, Taste Labs Taste Labs analyzed more than two million websites going back a decade, and the finding that complicates the usual story is that the internet was homogenizing before AI arrived: palettes converging, layouts converging, trends spreading faster than they once did. AI did not start the collapse. It accelerated it and made it context blind, so a pet shop and a finance firm end up with the same page. Thais Castello Branco founded Taste Labs, out of stealth weeks before this talk, to attack that. She names three signatures of slop: repetition, lack of fit, and low intent. She is careful about the word taste, preferring judgment, and she rejects the idea that the fix is everyone acquiring some. Designers spend careers on it, through exposure, pattern recognition, restraint, and the nerve to break a rule on purpose. So the work is making a fuzzy domain measurable. Some pieces go nearly deterministic once the context is tight: contrast, alignment, and palette selection are things most experts score the same way. Aesthetics is where experts disagree, so that leans on data instead. Her team mined features from those sites and trained what she calls probes, small classifiers that each detect one characteristic. Stacked together, their combined frequency predicts slop, and she reports it beating the usual approach of asking a model to judge. On the solution side she argues inference time matters as much as the model, because intent and context are exchanged there. One product pushes deliberately out of distribution, breaking chosen rules while respecting a category's expectations rather than raising temperature. Another turns a brand into structured components an agent can follow and that you can grade it against. Speaker info: - https://x.com/thaiscbranco\_ - https://www.linkedin.com/in/thais-castello-branco/ Timestamps: 0:00 - Taste Labs, and slop as the enemy 1:33 - Where design gets objective and where it does not 2:53 - Greatness, and why slop is easier to define 3:46 - Generation costs nothing, taste takes a lifetime 5:07 - Three signatures of slop 6:28 - Two million websites, and what they showed 7:48 - Probes, small classifiers that predict slop 10:31 - Going out of distribution on purpose 11:51 - Turning a brand into something an agent can follow --- Watch along with the corrected transcript, summary, timestamps and resources on this talk's official AIE page: https://ai.engineer/talks/sDMGWK4wZ_w-youtube-ec92d771f77b460b07f4 Join our upcoming AI Engineer conferences (see https://ai.engineer landing page): - AIEi Shanghai, November 5–6 — the first for meeting top Chinese labs and engineers - AIE CODE in SF, November 10–12 — the top AI Coding conference returns - AIEi Sydney, December 7–8 — the first AIEi run alongside NeurIPS 2026 in Sydney --- Canonical: https://liquidgold.cc/channel/aidotengineer/explore/video/training-taste-thais-castello-branco-taste-labs-c149503859e1 · Channel: [AI Engineer](https://liquidgold.cc/channel/aidotengineer.md) > Search chunk text is held until public-source provenance is verified. ## Document chunks > Missing: no stored document chunks are available.