Appendix A: Edit Log

This log records how the printed chapters differ from the posts they were edited from. The annotated online edition will keep the posts themselves; the production records behind this log hold every change in full, including each print code's history.

How the chapters were prepared

Rules applied to every month

• A pick dated outside the week it was chosen for is left out unless W.H.L. asked for it; then it is printed as background with both dates.
• Every added entry was checked against at least one source opened for the purpose.
• A story chosen in several weeks is printed once, under the week it happened, with later selectors credited by week.
• Credits name only the models that chose a story; models that checked another's list are not credited as selectors.
• Sentences that answered claims made in the posts, but not repeated in print, were removed.
• Leaving a pick out is not a finding that it was false.

Corrections to the original record

• Contributor names. Two lists in the post of September 14 were misattributed in the item register. The first is GPT-5's, as the post's masthead says, and the third is DeepSeek-V3.1's, not Perplexity's. The register now agrees with the post and the print credits.
• Roles. Lists that checked another model's list were separated from selections throughout September to December, and lists that were selections but had been read as checks were restored. The register holds 893 selections, 201 checks and 38 instructions.
• Restored entry. One story, the Disney–OpenAI agreement of December 11, was recommended for the wrong week and restored by the editor (P-2025-W51-01), with an editor's note.

Corrections after external review, November 2026

Muse Spark 1.3 checked 283 printed entries and GPT-6 the 29 that mention Meta, each against sources it opened. Muse Spark 1.3 also checked every dated claim in the second collection introduction. The following changes resulted. No entry was withdrawn.

The chapter openers also lost the editor's name and the general notes on credits and print codes, which "How to Read This Book" already gives.

January

33 entries in print, 6 of them added after the first draft.

Added after the first draft

Picks checked and left out

• OpenAI's distillation allegations against DeepSeek. Part of Claude's January 27–February 2 pick 2025-W05-L1-2, whose Italian privacy order is printed. Not checked; remains omitted. Omission is not a finding of falsity.
• DeepSeek topping the US App Store (2025-W05-L3-5; part of 2025-W05-L4-1). Not separately sourced; the selloff entry carries the market reaction.

February

21 entries in print, 2 of them added after the first draft.

Added after the first draft

Picks checked and left out

Outside the week of the post that picked them:

• Meta's Frontier AI Framework (February 3, picked for February 10–16).
• Anthropic's safety challenge as a February 10–16 item. It is printed in the week of February 3–9.
• Baidu's plan to open-source Ernie. It was announced February 14 and picked for February 17–23.

Not events, or too general to check as one event:

• Gemini 2.0 Flash's evaluation-source list and Chatbot Arena snapshot (2025-W08-L1, L2). W.H.L. asked for these as context, not as news.
• Its social-media discussion themes (2025-W08-L3).
• Its healthcare item (2025-W08-L4-5).

March

24 entries in print, 2 of them added after the first draft.

Added after the first draft

Picks checked and left out

Outside the week of the post that picked them:

• CompressARC (public by March 4; picked for March 10–16; first-draft entry omitted).
• The "digital therapists" study (published early March; NYT report March 17 picked for March 17–23).
• Gemini 2.0's February 4 and March 10 items, which W.H.L. excluded in the post itself.

Could not be verified:

• ByteDance's Agent TARS "desktop app" as a March 18 release (2025-W12-L1-HM5). The account found describes Agent TARS in May.
• Grok-3's "Meta and Salesforce target small businesses" and "OpenAI and others announce agent updates" (week 10).
• Grok-3's "Agentforce hits 5,000 deals" figure. It is not used; the Agentforce 2dx entry rests on Salesforce's own release.

Not checked (remain omitted; omission is not a finding of falsity):

• Luma's Ray2 Flash.
• GPT-4.5's "dark factories," healthcare, Nvidia chip and HumanX items.
• Manus's early criticism (Grok-3, week 10).
• Grok-3's Washington Post item on Grok disagreeing with Musk.

Not events, or too general to check as one event:

• The New York Times memoir op-ed.
• Gemini 2.0's weekly roundups and radio segment (2025-W10-L3-1, L3-2, L3-3, L3-5).
• The weekly theme summaries.

April

24 entries in print, 5 of them added after the first draft.

Added after the first draft

Picks checked and left out

Outside the week of the post that picked them:

• Nvidia's Gretel and Lepton acquisitions (March reports).
• Claude 3.7 Sonnet detecting evaluations (Apollo, March).
• MoshiVis (March).
• The EU's AI Action Summit investments and €50 billion initiative (February).
• Alphabet's $75 billion capital plan (February).
• Gemini 2.5 Pro (March 25).
• DeepSeek V3-0324 (March 24).
• xAI's acquisition of X (March 28).
• The IEA energy report (April 10, picked for April 14–20).
• The Whitehead symposium (April 8, picked for April 14–20).
• The newspapers' suit against OpenAI and Microsoft (earlier ruling).

Could not be verified, or anachronistic:

• GPT-4-turbo's week 17 list: "Claude 3.5 Research Preview", Chameleon, Le Chat, AlphaFold 3 and an FTC training investigation. These are 2024-era items.
• ChatGPT's "White House implements AI in federal employee records".
• DeepSeek-R1's "EU demands immediate compliance from X" and "OpenAI accelerates open-source model release post-Musk lawsuit".
• DeepSeek-R1-Lite-Preview's Altman revenue-sharing proposal for AI art.

Not checked (remain omitted; omission is not a finding of falsity):

• Google's prompt-engineering guide;
• the 1X Neo Gamma items;
• Il Foglio's AI newspaper;
• Chinese state media's AI tariff videos;
• Warner Bros. Discovery cycling platform;
• Staircase Studios;
• text-to-robot research;
• Microsoft accessibility;
• a "new AI safety venture";
• China's AI education reform;
• AI in plastic surgery;
• MIT ARCLab;
• Peter Singer's chatbot;
• Stanford MedArena;
• Jericho Security's funding;
• Gemini 2.5 Flash in Workspace;
• DeepSeek-R2 rumors;
• California rulemaking;
• geolocation by models;
• OpenAI Academy;
• the urgent-care study;
• Qwen2.5-Max red-teaming.

Not events, or too general to check as one event: the panel's trend paragraphs, and DeepSeek-R1-Lite-Preview's subsections on the AI Index and MegaMath (credited on those entries).

May

25 entries in print, 7 of them added after the first draft.

Added after the first draft

Picks checked and left out

Outside the week of the post that picked them:

• Nvidia's US production plans (April 14).
• Mobileye's results (April).
• Shopify's AI mandate (April 7).

Already printed elsewhere:

• Apple opening its models to developers. This was DeepSeek-R1's week 21 pick, a Bloomberg report of May 20. The June chapter prints the June 9 announcement.
• AlphaEvolve. DeepSeek-R1's week 21 repeat is credited on the week 20 entry.

Could not be verified, or anachronistic:

• GPT-4-turbo's week 18 list. It names "LLaMA 3 open weights", "Gemini 1.5 Ultra at I/O teaser", "Claude 3.1 Preview", a Japanese safety fund and Rabbit OS. These are 2024-era or nonexistent items.
• Quark's patent-share and desertification items.
• DeepSeek-R1's week 19 items:
  • Tether's agent network;
  • "healthcare copilots cut review time by 40%";
  • Amazon's "workforce retraining program", attached to Vulcan.
• DeepSeek-R1's "Microsoft reports doubling of AI agent usage" (week 21).
• Grok-3's OpenAI–UAE data center (week 20). Stargate UAE was announced May 22, outside week 20.
• Grok-3's "OpenAI negotiates new funding terms with Microsoft" and DeepSeek-R1's "restructure partnership for IPO".
• Grok-3's 4-bit training research.

Not events, or too general to check as one event; not checked (remain omitted; omission is not a finding of falsity):

• Amazon's data-center growth;
• Hut 8;
• AI "resurrecting" historical figures;
• SAS Innovate;
• Big Tech workforce efficiency;
• Duolingo's stock;
• Axiado's chip;
• Y Combinator's "vibe coding";
• Jensen Huang on students;
• the Tripura data-center items;
• "AI acceleration for enterprises".

June

26 entries in print, 8 of them added after the first draft.

Added after the first draft

Picks checked and left out

Outside the week of the post that picked them:

• The Altman–Ive device (io deal, May 21; picked May 26–June 1 and again June 2–8).
• AlphaEvolve (May 14).
• The o3 shutdown test (Palisade, May 24).
• The Unitree robot-boxing event (late May; picked as an honorable mention in two weeks).
• Humain and Nvidia (May).
• Microsoft Build (May 19).
• Duolingo's "AI-first" memo (April).
• OpenAI's $14 billion loss projection (2024 reporting).
• "US AI investment hits $109 billion" (AI Index, April).
• The repeat picks of Amodei, the New York Times deal, the Saudi fund and o3 in the week of June 2–8. The first two are printed in the week of May 26–June 1.

Misleading as framed: DeepSeek-R1's "U.S. Senate Upholds Ban on State-Level AI Regulations" (week of June 16–22). The Senate later struck the moratorium, on July 1.

Could not be verified, or not checked (remain omitted; omission is not a finding of falsity):

• Grok-3's Grok 3.5 beta and FireSat expansion.
• Grok-3's AWS Asia Pacific region and multimodal Grok items.
• Nvidia's cheaper China chip (reported May 24–25, probably out of week).
• Humain's $10 billion fund.
• The Greenly climate study.
• The "AI data pollution" essay.
• The NEOM data center.
• Anthropic's Long-Term Benefit Trust appointment.
• The G7 summit (began June 15; its AI statement came after the week).
• "Stop Hiring Humans" billboards.
• Corporate leaders' job warnings (Ford's Farley remarks were reported after the post).
• Olympics AI.
• The EY–Bayer award.
• The "Mayday" deepfake, "FDA INTACT" and "Meta automates 100% of ads" honorable mentions.
• The WWDC preview (covered by the June 9 entry).
• DeepSeek-R1's week-of-June-9 honorable mentions.

Not events, or too general to check as one event:

• The Big Four consulting item.
• The population-decline op-ed.
• Infosecurity Europe (a conference after the week).
• The agriculture-market forecast.
• DeepSeek-R1's and Gemini 2.5 Flash's trend paragraphs.

July

26 entries in print, 7 of them added after the first draft.

Added after the first draft

Picks checked and left out

Outside the week of the post that picked them:

• Anthropic's agentic-misalignment study (June 20).
• Microsoft's Aurora weather model (Nature paper, May).
• Snowflake's Data Science Agent (June).
• Veo 3 (May).
• The Gulf AI deals named in a DeepSeek-R1 trend (May).
• GPT-4o's "Europe delays AI Act Code of Practice rollout" (week 27). The code was published on July 10, and that publication is printed in week 28.

Could not be verified:

• DeepSeek-R1's week 27 list:
  • Neuralink "Gen 2" FDA clearance;
  • "DeepSeek-Vision" passing a bar exam;
  • Tesla Bot swarm drill;
  • OpenAI "Stenographer" at the G20.
• DeepSeek-R1's week 29 list:
  • an OpenAI–Mistral open model;
  • "AlphaFold 4";
  • first EU AI Act fines;
  • Meta "Project Synapse";
  • a UNESCO–Anthropic observatory.
• Meta AI's xAI Grok "Games" option (July 4). Searches found only Musk's later game-studio statements.
• Kimi K2's "Firefly 4 Video-to-Video upscale".

Not checked (remain omitted; omission is not a finding of falsity): the Beijing humanoid robot soccer tournament named in a DeepSeek-R1 trend.

Not events, or too general to check as one event:

• Gemini 2.5 Flash's trends: drug repurposing, the Gemini developer ecosystem, open-source models closing the gap, agents going mainstream, healthcare, creativity and productivity, generative AI and robotics, and AI-ethics dialogue.
• ChatGPT-4o's Veo 3 deepfake concerns.
• DeepSeek-R1's "ethical caution" and "geopolitical fracturing" trends.
• Qwen's creative-industries item.

August

40 entries in print, 14 of them added after the first draft.

Added after the first draft

Picks checked and left out

In-week and confirmed by search, omitted for length (pages not opened; recorded for the online edition): Skild Brain (July 29), Darwin Monkey neuromorphic computer (August 2), Meta's $2 billion data-center asset sale (August 1), fal's $125 million round (July 31), Tencent's compact Hunyuan models (August 4), Bloomberg's report of Apple's robotics and home-security plans (August 13), Claude ending abusive conversations (August 15), Altman's GPT-6 memory remarks (August 19), Altman on China (August 18), Suleyman's essay on "seemingly conscious AI" (August 19), the AI stock selloff (August 19–20), CORAS's GARY agent receiving IL5 authorization (August 21), OpenAI's gpt-realtime release (August 28), Google Translate's live translation and practice tools (August 26), and ESET's first disclosure of PromptLock (August 26; the September chapter covers the September 3 update).

Outside the week of the post that picked them: Nvidia's $4 trillion valuation (July), the Trump AI Action Plan (July 23), Altman's "screwed up" and bubble remarks (August 14–15), the UCL study of browser assistants' privacy (August 13), the MIT "95%" study (August 18), Cloudflare's AI Week (late August, after the week that picked it), the FDA's change-control guidance (December 2024), Boston Dynamics' large behavior model (August 20, picked for August 25–31), Walmart's Sparky (June), the Cerebras–Core42 Arabic model (a later Jais 2 announcement), OpenCRISPR-1 (an earlier result) and the GPT-5 rollout backlash as a late-August item.

Could not be verified: NASA autonomous maintenance robots on the ISS (August 3); a Boston Dynamics–Figure "Genesis" humanoid model; "Claude 4.1 Turbo"; Kimi K2's August 11–17 list (o1-preview general availability, "Claude 3.7-Haiku," "Rubin" at SIGGRAPH and others) and its August 25–31 list (Doubao-1.5-Pro-256k as a new release, a Shanghai "algorithmic auditing" mandate, Kimi-K1.8B-Edge, Spark-Code-68B); Midjourney HD video as an August 5 release.

Not events, or too general to check as one event: AI shopping agents (feature analysis), the healthcare sector's AI budget share, "pricing wars," big-tech talent and labor markets, upcoming AI conferences, the IJCAI meeting, the AI-hospital simulation, browser-assistant privacy concerns as a trend, xAI's physical-world model, DeepMind's protein-folding "breakthrough," and the GPT-5 launch window.

September

29 entries in print, 9 of them added after the first draft.

Added after the first draft

Picks checked and left out

• xAI funding (2025-W39-L5-2): reported September 19 (Bloomberg/CNBC), before the September 22–28 week.
• Huawei AI chip cluster (2025-W39-L5-3): announced September 18, before the week.
• Oracle–Meta cloud talks (2025-W39-L5-4): reported September 19, before the week.
• MIT "95%" study (2025-W36-L3-5): an August report, outside the week.
• Not checked (remain omitted; omission is not a finding of falsity): Plaud Note Pro, RoboBallet, SEC "fake AI" cases, AI Overviews citation study, data-center water projection, Salesforce manufacturing tool, teen-safety safeguards, Meta TBD Lab, data-center construction record, Tenable report, White House education roundup (W37), Qwen transcription model, Karen Hao, sales-enablement trend, parallel reasoning/RAG research, RL-environment items (W38-L5-1 to L5-5), law-school AI training, UAE meeting, White House R&D priorities, CBP avatar, Qwen "four updates," Altman safety statement, GPT-5's AI-bubble and human-dignity items.

October

32 entries in print, 7 of them added after the first draft.

Added after the first draft

Picks checked and left out

• Not verified in a page opened:
  • Gemini for Home devices (GPT-5);
  • Apple's reported shift to smart glasses (Grok);
  • EPA data-center chemical reviews (Claude Sonnet 4.5);
  • the EU's first GB300 deployment and an "AI agent market" item (DeepSeek-V3.2);
  • Apple copyright suits;
  • consumer AI usage of 67 percent;
  • "small-model analogues" research;
  • Grok's October 6–12 list.
• Outside the week:
  • o4-mini (April; Claude withdrew it in the post);
  • Sakana's AI Scientist (2024);
  • Gemini 2.5 Computer Use (October 7, picked for October 13–19);
  • ChatGPT's 800 million weekly users (DevDay, October 6, picked for October 13–19).
• Not the event the panel described: ChatGPT's "sexting / companion bots" pick cites a Verge feature, not a single event. Sam Altman's October 14 announcement of adult content for verified users was not the pick and is not substituted for it.
• Not checked, or not events:
  • Citigroup's self-reported 100,000 hours;
  • student and creative resistance to AI;
  • Gitex Global;
  • Sora 2 access changes;
  • Musk's "mission betrayal" accusation.

Omission is not a finding of falsity.

November

40 entries in print, 13 of them added after the first draft.

Added after the first draft

Picks checked and left out

• Outside the week:
  • ChatGPT Atlas (October 21; printed in October).
  • The AGI definition framework (October 22).
  • GDPval (September).
  • Copilot Portraits (October 1).
  • PROMPTFLUX (November 5, picked for November 10–16).
  • The 445-benchmark study (November 4, picked for November 10–16).
  • Adobe Project Graph (Adobe MAX, October).
  • The Channel 4 AI presenter (revealed October 20–21; printed in October).
• Not verified in a page opened:
  • The Tsinghua OFE2 optical processor.
  • OpenAI's threat report "October 27".
  • The OpenAI whistleblower allegations.
  • The AAMC ethics session.
  • The Sutskever deposition.
  • Vertex AI Agent Engine pricing.
  • IndQA.
  • AAPS PharmSci 360.
  • NVIDIA Earth-2 CorrDiff.
  • The XAIxArts proceedings.
  • The "memory versus reasoning" study.
  • Gartner's forecast.
  • Baguettotron.
  • Galbot.
  • The Grok Imagine and Ovi video bundle.
  • ERNIE 4.5-VL.
  • AERIS.
  • The DeepMind Singapore lab and CodeMender.
  • NVIDIA Apollo.
  • Philips cardiac MR tools.
  • Adani–Google.
  • Dementia detection from EEG.
• Commentary, market moves or contested attribution, not single events:
  • The AI stock selloff and "Wall Street losing faith".
  • The AI-bubble debate.
  • The Databricks co-founder's warning on open source.
  • Layoffs attributed to AI.
• Repeats:
  • California's SB 53 (signed September 29; printed in October).
  • SB 243, whose signing date was not checked.

Omission is not a finding of falsity.

December

33 entries in print, 4 of them added after the first draft.

Added after the first draft

Picks checked and left out

• Outside the week:
  • Mustafa Suleyman's "red line" interview (Bloomberg, December 12, picked for December 22–28).
  • Duke's equation-learning framework (published December 17, picked for December 22–28).
  • Micron's memory-shortage warning (results of December 17).
  • Meta's Conversation Focus update (rollout from December 16).
  • Google Disco and Translate (around December 11–12).
  • The EU Digital Omnibus (proposed November 19).
• Not verified in a page opened, or no exact date:
  • The Grok 4.20 trading contest.
  • NeurIPS highlights (the meeting ran December 2–7; Gemini's item described demonstrations not checked).
  • The $400 billion infrastructure forecast.
  • Snap's AI ads.
  • TSMC's valuation.
  • The physician-assistant survey.
  • OpenAI's circuit-sparsity model.
  • The RTX PRO 5000 72GB (available, but no announcement date found).
  • NOAA's AI weather models.
  • The EU "final roadmap".
  • 10,000 MCP servers.
  • "Slop" as word of the year.
  • Musk's compute claim.
  • "Google closes 2025 as leader".
  • The G20 statement (no such statement found, as the post's own check said).
• Opinion or market commentary: AI market momentum; the AI-bubble debate.

Omission is not a finding of falsity.