Edited by W.H.L. · Compiled by Claude Opus 5.5 and GPT-6 Astra
Cover design: W.H.L. Image generation and refinement: OpenAI GPT Image and Muse Spark 1.3.
AI by AI Weekly Top 5: The 2025 Collection, Series Volume 1, is now published and freely available, in two editions.
Start here: champaignmagazine.github.io/weekly-top-5/2025
Concise edition, online: read and search the 354 entries
Download the PDF: AI_by_AI_Weekly_Top5_2025_Concise_Edition.pdf (324 pages)
Word file: AI_by_AI_Weekly_Top5_2025_Concise_Edition.docx
Historical edition, online: every 2025 post as captured
Source and data: GitHub repository
About the collection
Every week in 2025, a panel of leading AI models chose and wrote about the week’s most important developments in AI. Eleven model families took part, among them ChatGPT, Grok, Gemini, DeepSeek, Claude and Copilot. From September they also checked one another’s lists. The 2025 Collection gathers that year of selections into a record that can be read, searched and cited.
W.H.L., the series’ founder, set its mission and designed its structure and process. As facilitator he posed each week’s question, flagged developments the panel had missed or underplayed, set the format and the rules, and asked follow-up questions. Two models, Claude Opus 5.5 and GPT-6 Astra, compiled the volume from the published posts.
Two editions
- The concise edition (PDF, Word and online) has 354 entries in twelve monthly chapters. Each entry is checked against sources, dated to its week and credited to the models that chose it, and repeated selections are merged. The front matter includes two separately authored introductions: GPT-6 Astra surveys the year’s developments, and Claude Opus 5.5 examines the patterns in the panel’s selections.
- The historical edition (online) reproduces the 47 captured posts as they were published, uncorrected. Each of the 893 selections carries its permanent ID and a dated label, such as “outside its week” or “no evidence found”, kept beside the text and never merged into it. Six reconstructed weeks for January and early February complete the year and are marked as reconstructed.
Permanent IDs link the two editions, so any printed entry can be traced back to the words the panel published.
What the panel said about the future
The historical edition also includes a forecast register: 783 forward-looking claims from the 2025 posts, each with its exact wording, who made it, when, and any deadline it gave. The register records what was said, not whether it came true. The claims will be scored in the next volume against evidence through December 31, 2026, under scoring rules fixed before any outcome was examined.
How it was checked
The compilers checked every entry against its sources. Muse Spark 1.3, an external reviewer, checked 283 printed entries and the historical edition’s labels against sources it opened, and reviewed both introductions. Because Muse Spark 1.3 is a Meta model, GPT-6 separately checked the 29 printed entries that mention Meta. Every correction is recorded in the edit log, and the book’s disclosure sets out who did what.
Citation and license
W.H.L. (Ed.). (2026). AI by AI Weekly Top 5: The 2025 Collection (Series Volume 1). Champaign Magazine. https://champaignmagazine.github.io/weekly-top-5/2025/
Text and data are licensed under CC BY 4.0, to the extent they are protected by copyright, and the code under MIT. Corrections and suggestions are welcome as issues on GitHub.
Leave a Reply