AI Content Disclosure
Last updated 2026-04-15
AllScience believes in full transparency about how content on our platform is created. This page explains our AI-assisted writing process, the quality standards we enforce, and the models we use.
1. Our AI Writing Process
Books and educational content published through our platform are written using Qwen3, our own fine-tuned large language model. This is not a generic chatbot or a third-party API—it is a model we have trained specifically for nonfiction writing using Direct Preference Optimization (DPO) and Group Relative Policy Optimization (GRPO) on thousands of human-reviewed quality pairs.
Every piece of content goes through multiple revision cycles. The model generates a draft, a separate quality evaluation scores it against our rubric, and if it falls below threshold the content is revised automatically. This loop continues until the piece meets our standards or is flagged for human editorial review.
2. The 9.5 Quality Gate
Every piece of content must score 9.5 out of 10 across five dimensions before it is published:
- Grammar — checked against 430+ rules for spelling, syntax, punctuation, and style consistency.
- Clarity — sentence structure, logical flow, and readability appropriate to the target age group.
- Engagement — narrative pacing, reader interest, and age-appropriate tone.
- Accuracy — factual correctness verified against source material and domain knowledge.
- Originality — statistical AI-detection analysis and plagiarism screening to ensure unique content.
Content that scores below 9.5 is rejected and either re-generated or sent for manual editorial intervention. For full details on our content standards, see our AI Content Policy.
3. What Models We Use
Qwen3 (Text Generation)
Our fine-tuned version of the Qwen3 large language model, trained on human-reviewed nonfiction writing pairs. Handles book drafting, revision, and content generation.
Kokoro (Text-to-Speech)
Neural TTS engine used for audiobook narration. Produces natural, expressive speech with per-chapter mastering via FFmpeg.
FLUX (Image Generation)
AI image generation model with our custom LoRA fine-tunes for book cover art. Trained on curated datasets for consistent visual style across our catalog.
All models run on our own GPU infrastructure. We do not send your data or our content through third-party AI APIs. Every model runs on hardware we control, which means zero external data sharing for content generation.
4. AI vs. Human
All AI-generated content on our platform is clearly labeled as such. We do not misrepresent AI-written work as human-authored.
Every piece of content receives human editorial review before publication. Our editorial team reviews quality scores, spot-checks content for accuracy and tone, and makes final publish/reject decisions.
Author names listed on books are pen names managed by our editorial team. This is disclosed here and in our Terms of Service. The writing itself is AI-generated, quality-gated, and editorially supervised.
5. How to Report Issues
If you find a factual error, quality issue, or any concern with content on our platform, we want to hear about it. Please contact us with details about the specific content and the nature of the issue.
All quality reports are reviewed within 48 hours. If the issue is confirmed, the content is pulled, corrected, and re-published only after passing our quality gate again.