# content-trust-series

Published articles for content-trust-series.

This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles.

## A Repeatable Human-in-the-Loop Process for Large-Scale LLM Classification

DevFeed: [A Repeatable Human-in-the-Loop Process for Large-Scale LLM Classification](<https://devfeed.tech/articles/stop-building-models-start-building-systems-22564.md>)

Original publisher: [Read original article](<https://tech.scribd.com/blog/2026/fast-llm-human-in-the-loop-classification.html>)

Author: Anish Kumar

Published: 2026-07-11T00:00:00Z

Content type: article

Language: en

Sources: [Scribd Tech](<https://devfeed.tech/sources/scribd-tech.md>)

Topics: [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Prompt Engineering](<https://devfeed.tech/topics/prompt-engineering.md>), [Inference](<https://devfeed.tech/topics/inference.md>), [datasets](<https://devfeed.tech/topics/datasets.md>), [data](<https://devfeed.tech/topics/data.md>)

Tags: [asynchronous](<https://devfeed.tech/tags/asynchronous.md>), [batch](<https://devfeed.tech/tags/batch.md>), [content-trust-series](<https://devfeed.tech/tags/content-trust-series.md>), [cost](<https://devfeed.tech/tags/cost.md>), [data](<https://devfeed.tech/tags/data.md>), [databricks](<https://devfeed.tech/tags/databricks.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [featured](<https://devfeed.tech/tags/featured.md>), [inference](<https://devfeed.tech/tags/inference.md>), [llm](<https://devfeed.tech/tags/llm.md>), [machinelearning](<https://devfeed.tech/tags/machinelearning.md>), [models](<https://devfeed.tech/tags/models.md>), [production](<https://devfeed.tech/tags/production.md>), [prompt](<https://devfeed.tech/tags/prompt.md>), [scribd](<https://devfeed.tech/tags/scribd.md>), [workflow](<https://devfeed.tech/tags/workflow.md>)

### AI overview

The article presents a repeatable human-in-the-loop process for large-scale LLM classification. It combines fast-model labeling, judge-model disagreement detection, targeted SME review, a golden dataset built from corrections, and selective prompt iteration.

### Source excerpt

LLM models change. Prompt quality changes. Cost changes. We assumed that from day one.

## Dual-Embedding Trust Scoring

DevFeed: [Dual-Embedding Trust Scoring](<https://devfeed.tech/articles/dual-embedding-trust-scoring-22563.md>)

Original publisher: [Read original article](<https://tech.scribd.com/blog/2026/content-trust-score.html>)

Author: Eric Chang

Published: 2026-02-25T00:00:00Z

Content type: article

Language: en

Sources: [Scribd Tech](<https://devfeed.tech/sources/scribd-tech.md>)

Topics: [trust and safety](<https://devfeed.tech/topics/trust-and-safety.md>), [Embeddings](<https://devfeed.tech/topics/embeddings.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [data](<https://devfeed.tech/topics/data.md>), [datasets](<https://devfeed.tech/topics/datasets.md>), [genai](<https://devfeed.tech/topics/genai.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [content-trust-series](<https://devfeed.tech/tags/content-trust-series.md>), [data](<https://devfeed.tech/tags/data.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [embeddings](<https://devfeed.tech/tags/embeddings.md>), [featured](<https://devfeed.tech/tags/featured.md>), [generative-ai](<https://devfeed.tech/tags/generative-ai.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [machinelearning](<https://devfeed.tech/tags/machinelearning.md>), [pii](<https://devfeed.tech/tags/pii.md>), [privacy](<https://devfeed.tech/tags/privacy.md>), [research](<https://devfeed.tech/tags/research.md>), [scribd](<https://devfeed.tech/tags/scribd.md>), [trust](<https://devfeed.tech/tags/trust.md>), [trust-and-safety](<https://devfeed.tech/tags/trust-and-safety.md>)

### AI overview

Scribd describes a Content Trust Score that combines Generative AI signals, proprietary multilingual embeddings, and classical machine learning to assess the severity of documents violating defined trust and safety pillars. The research covers illegal, explicit, privacy/PII, and low-quality content, using annotated data from roughly 100,000 documents.

### Source excerpt

Scribd is a digital library serving academics and lifelong learners, offering hundreds of millions of documents. This very nature presents a significant concern: content trust and safety. Protecting our library from undesirable and unsafe content is a top priority, but the multilingual and multimodal (text and images) nature of our platform makes this mission very challenging. Also, while third-party tools exist, they often fall short, lacking the nuance to handle our specific trust and safety categories.

## Deploying a Cost-Effective, Scalable PhotoDNA System for CSAM Detection

DevFeed: [Deploying a Cost-Effective, Scalable PhotoDNA System for CSAM Detection](<https://devfeed.tech/articles/deploying-a-cost-effective-scalable-photodna-system-for-csam-detection-22565.md>)

Original publisher: [Read original article](<https://tech.scribd.com/blog/2026/photodna-csam-detection.html>)

Author: Anish Kumar

Published: 2026-01-20T00:00:00Z

Content type: article

Language: en

Sources: [Scribd Tech](<https://devfeed.tech/sources/scribd-tech.md>)

Topics: [Perceptual hashing](<https://devfeed.tech/topics/perceptual-hashing.md>), [observability](<https://devfeed.tech/topics/observability.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Microsoft](<https://devfeed.tech/topics/microsoft.md>)

Tags: [auditability](<https://devfeed.tech/tags/auditability.md>), [aws](<https://devfeed.tech/tags/aws.md>), [content-trust-series](<https://devfeed.tech/tags/content-trust-series.md>), [cost](<https://devfeed.tech/tags/cost.md>), [databricks](<https://devfeed.tech/tags/databricks.md>), [false-positives](<https://devfeed.tech/tags/false-positives.md>), [featured](<https://devfeed.tech/tags/featured.md>), [hashing](<https://devfeed.tech/tags/hashing.md>), [lambda](<https://devfeed.tech/tags/lambda.md>), [latency](<https://devfeed.tech/tags/latency.md>), [legal](<https://devfeed.tech/tags/legal.md>), [microsoft](<https://devfeed.tech/tags/microsoft.md>), [observability](<https://devfeed.tech/tags/observability.md>), [privacy](<https://devfeed.tech/tags/privacy.md>), [reporting](<https://devfeed.tech/tags/reporting.md>), [scale](<https://devfeed.tech/tags/scale.md>)

### AI overview

Scribd describes a production system for detecting known Child Sexual Abuse Material using PhotoDNA perceptual hashes. The article covers its integration with NCMEC reporting, ingestion-scale architecture, operational model, cost considerations, and safety-related constraints.

### Source excerpt

Child safety is a non-negotiable responsibility for any platform that hosts user-generated content. Over the last year, we designed and deployed a production system that detects known Child Sexual Abuse Material (CSAM) using PhotoDNA perceptual hashes, integrates with the National Center for Missing and Exploted Children's (NCMEC) reporting system, and scales efficiently across our ingestion surfaces. This post explains the problem we set out to solve, how PhotoDNA hashing works, the online child-protection ecosystem (NCMEC, Tech Coalition, Project Lantern), our architecture and operational model, cost considerations, and key learnings.