About Shift
Shift (？) is a sourced, bilingual AI news site maintained by orocoa. The public site is served at the independent domain root. Chinese and English share stable event IDs, and every story ends with links to its original sources.
Author
The author is orocoa (email orocoa@kai-nex.com, GitHub github.com/orocoa). The same contacts appear in the site’s Contact dialog.
Sources and filtering
On a fixed Asia/Shanghai schedule the site incrementally collects headlines, dates and original links from configured sources: QbitAI, Techmeme, the GitHub Changelog (AI-related only), The Decoder, OpenAI, Google DeepMind and the Hugging Face Blog; large official feeds are typically capped at the latest 10 items.
The first run each week also checks Last Week in AI and Hugging Face Papers for gaps.
Leads are de-duplicated, then compared as a group on substantive impact on AI direction, productivity and tool choice and on new information; there is no quota by count, brand or score.
Selected events are reported from primary material—announcements, docs, pricing, model cards, papers or repositories. Vendor claims are attributed; unknown dates are not guessed. See Data sources and methods for the full list.
Updates and refresh
Editorial rounds follow the existing Asia/Shanghai schedule (including the noon Kitchen pass). Stories publish only after independent ShiftQA review and approval.
When you return to the tab, the app conditionally re-requests feed-index with an ETag: unchanged responses transfer nothing; updates merge by stable event ID while preserving scroll position, the open article, language and UI state—without a full page reload.
Original sources
Each story uses numbered citations in the body and ends with a de-duplicated list of original source links. Crawlable static article pages include the same source list and a path back into the live reader.
Weekly selection and archive
The site offers Day stories, a Weekly selection and a Models board. Day pieces archive by real publication date. Weekly issues cover completed Shanghai calendar weeks after a separate whole-issue review; day approval does not automatically place a story in the week.
How to read
The homepage is a single-page app: switch Chinese/English in the header, move among Today / This week / Models, open a card for detail, and close it to restore the prior list position.
Crawlable static article and topic pages, plus this About page and Data sources and methods, are also available for search engines and non-script clients.
Limits
Coverage is not exhaustive; discovery and publication depend on the sources above, editorial selection and review cadence.
The site does not claim user counts, rankings or marketing efficacy. The Models board follows the current upstream rules, keeps the last complete snapshot on failure, and never gates news selection by model scores.
Public descriptions stick to the current operating configuration and published content; unpublished processes or results are not invented.