# @ai_database: X profile (context) **Lists:** [[Users/@ai_database/Chronological List|Chronological list]] · [[Users/@ai_database/Topical List|Topical list]] **Post analyzed:** [[Users/@ai_database/Post Analyses/Post Analysis - ai_database on Swarm Telepathy KV-Cache Transfer (2026-10-28)|MIT "telepathy" between AI models (Oct 28, 2026)]]: 🟢/🟡 accurate summary of a real preprint; "telepathy" and the headline numbers need caveats **Topic:** [[Topics/LLMs/LLMs|LLMs]] > [!important] Interaction with [[marshallpittman]] > - Marshall bookmarked the "telepathy" post. No replies, quotes or mentions between @ai_database and @marshallpittman were found in AIDB's 40 timeline items, the 24 sampled quotes, or Marshall's recent posts. Likes cannot be checked. ## Identity - **Display name:** AIDB · **Handle:** [@ai_database](https://x.com/ai_database) · **User ID:** 1119122275033161730 - **Bio (verbatim, Japanese):** "論文ベースでAIのノウハウと科学技術にキャッチアップしています。運営8年目のウェブサービスAIDBの公式Xアカウントです。" (My translation: "We catch up on AI know-how and science and technology based on research papers. Official X account of the web service AIDB, now in its 8th year.") - **Joined:** Apr 19, 2019 · **Location:** Tokyo, Japan · **Verified:** business (organisation) check - **What AIDB is:** a Japanese web service that summarises AI research papers in plain Japanese. The operator's company and staff were not researched ⚠️. ## Stats (X API, Oct 30, 2026) | Followers | Following | Posts (lifetime) | Listed | Media | Likes given | |---:|---:|---:|---:|---:|---:| | 51,486 | 119 | 10,579 | 863 | 1,709 | 12,589 | ## What it posts about From the [[Users/@ai_database/Topical List|Topical List]] (40 items): | Theme | Items | Likes | Views | |---|---:|---:|---:| | Research results: games, physics, science, vision | 17 | 3,585 | 646,796 | | Practical use: prompting, Claude Code, AI search | 14 | 604 | 81,428 | | AI-to-AI communication and agents | 9 | 3,006 | 464,390 | 1. **Research results (17 items).** Paper summaries: AI agents built by "GPT-6 Astra" and "Opus 5.5" topping a Hearthstone Battlegrounds ranked server (Oct 26, 1,588 likes, 441K views), an AI that noticed altered physics laws in a simulated world (Oct 28, 1,420 likes), a red-heart emoji changing coding solutions (Oct 30), vision-language models. 2. **Practical use (14).** Prompting tips, CLAUDE.md and Claude Code habits, how AI search engines choose citations, summarising and context compression. 3. **AI-to-AI communication and agents (9).** The analyzed "telepathy" post (2,867 likes, 412K views) and a repost of @MTSlive's "neuralese" explainer. **Style.** Each post is a plain-Japanese summary of one paper, usually with a figure, followed by self-replies with "関連記事" (related article) and "論文：" (paper) links. 22 of the 40 items are such self-replies. Numbers are usually lifted straight from the paper's abstract; framing words ("telepathy", "noticed") are catchier than the papers'. ## Fact-check scorecard (this project) | Post | Verdict | |---|---| | [[Users/@ai_database/Post Analyses/Post Analysis - ai_database on Swarm Telepathy KV-Cache Transfer (2026-10-28)\|Oct 28: MIT "telepathy" between models]] | 🟢/🟡 faithful to arXiv:2610.15095; "telepathy" is the authors' own branding but means KV-cache transfer; 40%/4× hold only under the paper's compaction setup | ## Interactions with @marshallpittman See the box above. ## Gaps - 40 timeline items (Oct 25–30, 2026) of 10,579 posts; only one post fact-checked. - The company behind AIDB was not researched ⚠️. - Japanese text translated by me; no professional translation.