AI Policy & Editorial Team
SINGULISM uses AI (large language models) to assist with article production, while keeping accuracy and neutrality as a tech publication our top priority. This page explains exactly how we use AI and where human editors are involved.
Editorial Team
SINGULISM articles are produced by the team below. Each member's role and responsibility is clearly defined, making our hybrid AI + human editing process fully transparent.
SINGULISM AI Editorial Team
AIAI + Human Review
A hybrid editorial model where multiple large language models (MiMo / DeepSeek / GPT-4o) draft articles and human editors perform final fact-checking, structure refinement, and copyediting. To ensure transparency, AI involvement and the editorial workflow are disclosed on every article page.
SINGULISM Editor-in-Chief
HumanEditor-in-Chief & Fact-Checking Lead
Responsible for final review of AI-generated articles, publication decisions, and corrections. Performs cross-checking against multiple primary sources.
Article Production Workflow
- Information gathering (automated): We collect the latest topics from multiple trusted news sources (including TechCrunch, The Verge, NHK, Reuters, Wired, and Chinese-language tech outlets) via RSS and scraping. Non-news content (sales, jokes, best-of roundups) is automatically filtered out.
- First draft (AI): Multiple LLMs (MiMo, DeepSeek, GPT-4o) take turns writing a Japanese first draft, with automatic fallback. The prompt explicitly forbids fabricating numbers, statistics, or proper nouns not present in the source.
- Post-processing (automated): Simplified Chinese characters, Chinese patterns, typos, and translation garble are detected and replaced. We also check for incomplete articles and discard anything that doesn't meet our quality bar.
- English translation (AI): The Japanese article is translated into English using a translation chain that prioritizes the free tier (GitHub Models).
- Final review (human): The editor-in-chief fact-checks, refines structure, and copyedits. Factual errors or inaccuracies lead to rejection; we hold publication when in doubt.
- Publication & corrections: After publication we accept reader feedback and act on it quickly under our corrections policy.
Five Rules for AI Use
- Grounded in primary sources: The LLM is instructed to write only from the provided RSS/scraped content. Fabricating numbers, statistics, or proper nouns not present in the source is forbidden at the prompt level.
- No publication without human review: Every article is reviewed by a human editor before publication. Raw LLM output is never published as-is.
- Disclose AI involvement: Every article page and its JSON-LD expose whether AI assisted the writing in a machine-readable format, so readers can identify AI involvement.
- Cite sources: Every article carries a `source` and `sourceUrl` so readers can reach the original information.
- Discard subpar articles: Articles with Chinese contamination, mid-paragraph truncation, or abnormal repetition are destroyed rather than published. We prioritize quality over quantity.
Frequently Asked Questions
Why use AI at all?
The volume and pace of global tech news exceeds what a single human can cover. Using AI for first-draft generation lets our human editors spend more time on fact-checking and editorial judgment. AI writes the draft; humans guarantee the quality.
How do you prevent AI hallucinations?
We apply three layers of defense: (1) the prompt explicitly forbids inventing numbers not in the source, (2) post-processing detects and discards articles with Chinese contamination, abnormal characters, or truncation, and (3) the human editor fact-checks before publication. If incorrect information still slips through, our corrections policy requires us to act quickly with corrections or takedowns.
Will you stop using AI in the future?
There are no current plans to do so. However, we regularly review the scope of AI assistance (drafting, translation, copyediting) and the models we use, adjusting the balance between quality and transparency. Implementation details are available in the source code (public repository).
Corrections & Contact
Please use the contact form to report any article issues or request a correction. The editor-in-chief reviews every report and publishes a correction promptly when warranted.