AI BriefingStarting today, large-scale improvements related to feed quality and automation are being applied.
1. Fixed Bug Where the Same Article Remained "Processing"
There was a bug where the collection pipeline repeatedly reverted articles that had already been processed by AI back to the pending state. This was particularly noticeable with domestic corporate blogs (Toss, LINE, MyRealTrip, Olive Young, Musinsa, Yeogi Aettae, KT Cloud, Replit, etc.), where the "1" and "2" counts in the sidebar would not decrease no matter how many times AI processing was run.
After the fix:
- Articles that have been processed by AI are kept as is even when the next collection occurs
- Sidebar counts actually drop to 0
- All 12 accumulated articles from today were reprocessed to restore the current normal state
2. Automated AI Processing Schedule (Admin)
Previously, only "Auto Collection" could be registered as a schedule, and AI processing required someone to manually press a button after collection was complete. Now, AI processing is also registered as an independent cron schedule, so summarization and translation are automatically executed after collection is complete.
As a result, the time it takes for new articles to appear in the feed is significantly reduced (Previously: 1h collection + manual wait → Now: 1h collection + automated AI within ~15 minutes).
3. Natural Language Schedule Input (Admin)
When setting up schedules, instead of directly entering cron expressions like "0 9 * * *", you can now input Korean sentences such as "Every day at 9 AM", "Weekdays at 6 AM, Noon, and 6 PM", or "Every 30 minutes on weekends only". The AI automatically converts these and shows a preview; once confirmed and saved, it is registered in the crontab.
4. Enhanced Deduplication of Same Topics
The AI now groups multiple articles covering the same release, paper, or event during the collection stage and exposes only the single highest-scoring article in the feed. Combined with the secondary filter on the screen, this provides two layers of protection to ensure identical titles do not appear side by side.
5. External Crawling for Link Posts Without Body Text
Link posts from platforms like Reddit and Hacker News that have "only a title and no body text" now automatically crawl the connected external article pages to summarize based on the actual content. Posts where external content cannot be retrieved are automatically excluded from the feed.
6. GitHub Releases Summary Restoration
Due to the impact of the reset bug, several releases remained in the pending state despite having "Summaries available," causing them to disappear from the feed. Today, a bulk reprocessing was performed, and recent releases from the four repositories—pytorch, transformers, langchain, and ollama—are now visible in the feed with their summaries.
7. Improved Dark Mode Readability
The rendering of bold text in AI summaries is now displayed consistently in both light and dark modes.
These changes apply to all articles currently being collected, and previously mis-saved articles will be sequentially reflected during the reprocessing and recollection process.