Change Log
Every merge, in order. This is what makes the Handbook auditable rather than a black box — each row names the source it came from and where it landed.
| Date | Source ID | Section(s) | What was added |
|---|---|---|---|
| 15 Jul 2026 | DK-2 | AI Model Landscape; Big Tech, Legal & Business; Space, Robotics & Hardware | Initial merge — Grok 4.5/GPT-5.6/Muse Spark/Fable competitive landscape; Apple v. OpenAI lawsuit; SpaceX/China rocket milestones; 1X robotic hand; state AI regulation. |
| 15 Jul 2026 | DK-3 | AI Model Landscape; AI-Powered Content Creation | Initial merge — GPT-5.6 capability/cost claims; unified ChatGPT app; Sites instant deployment; GPT Live; competitor roundup. |
| 15 Jul 2026 | DK-4 | AI-Powered Content Creation | Initial merge — keyframe transition technique; Gemini Omni background edits; Remotion vs. diffusion; creator's 95/5 human/AI ratio. |
| 15 Jul 2026 | DK-5 | AI Model Landscape; AI-Powered Content Creation; Big Tech, Legal & Business | Initial merge — Fable 5 withdrawal/return; GPT-5.6 variants; BeautyBench; OpenAI government ownership proposal. |
| 15 Jul 2026 | DK-6 | AI Model Landscape; Big Tech, Legal & Business; Space, Robotics & Hardware; AI Ethics (new) | v2 merge — rumor-flagged Fable 5/Mythos 5 export-control account; Capybara/Numbat codename speculation; local AI hardware. |
| 15 Jul 2026 | DK-7 | AI-Powered Content Creation; AI Ethics (new) | v2 merge — HeyGen/ElevenLabs/Higgsfield pipeline; consent step; disclosure debate. |
| 15 Jul 2026 | DK-8 | AI Model Landscape; AI-Powered Content Creation | v2 merge — Seedream 5.0 Pro vs. GPT Image 2; localized editing; live web-integrated generation. |
| 15 Jul 2026 | DK-9 | AI Ethics (new) | v2 merge — multi-signal AI-video detection checklist; named misuse categories. |
| 15 Jul 2026 | DK-10 | AI Agentic Platforms (new) | v3 merge — ChatGPT Work autonomous execution, desktop file access, Sites, Scheduled Tasks. |
| 15 Jul 2026 | DK-11 | AI Agentic Platforms (new) | v3 merge — Claude Skills as reusable workflows; third-party skill security risks. |
| 15 Jul 2026 | DK-12 | AI-Powered Content Creation | v3 merge — NotebookLM Short Video Overviews; Nano Banana 2 Light. |
| 15 Jul 2026 | DK-13 | AI Agentic Platforms (new) | v3 merge — four-layer Claude ecosystem description. |
| 15 Jul 2026 | DK-14 | AI Agentic Platforms (new) | v3 merge — beginner Claude Code walkthrough. |
| 15 Jul 2026 | DK-15 | AI Agentic Platforms (new); General Operating Principles | v3 merge — ICC framework, context interview, hallucination safeguards. |
| 15 Jul 2026 | DK-16 | AI Agentic Platforms | v4 merge — Claude Code Artifacts live hosted apps. |
| 15 Jul 2026 | DK-17 | AI-Powered Content Creation; AI Agentic Platforms | v4 merge — Structure/Context/Templates/Skills framework. |
| 15 Jul 2026 | DK-18 | AI Model Landscape; AI Agentic Platforms | v4 merge — beginner three-tier model-selection framework. |
| 15 Jul 2026 | DK-19 | AI Ethics | v4 merge — self-learning AI-driven malware prototype research. |
| 15 Jul 2026 | DK-20 | Big Tech, Legal & Business | v4 merge — rumored 2026 hardware-maker product roadmap. |
| 15 Jul 2026 | DK-21 | AI Agentic Platforms | v4 merge — AI-as-teammate framing; realization-gap statistics. |
| 15 Jul 2026 | DK-22 | Space, Robotics & Hardware | v4 merge — human-centered robotics/interaction design. |
| 15 Jul 2026 | DK-23 | AI Agentic Platforms | v4 merge — counter-perspective on agent reliability. |
| 15 Jul 2026 | DK-24 | General Operating Principles | v4 merge — confidence-tiering discipline reinforcement. |
| 15 Jul 2026 | DK-25 | Space, Robotics & Hardware | v5 merge — UBTech commercial companion-humanoid product launch. |
| 15 Jul 2026 | DK-26 | Space, Robotics & Hardware | v5 merge — Physical AI/robot data gap concept. |
| 15 Jul 2026 | DK-27 | AI-Powered Content Creation | v5 merge — Midjourney feature roundup. |
| 15 Jul 2026 | DK-28 | AI-Powered Content Creation | v5 merge — full Midjourney parameter reference. |
| 15 Jul 2026 | DK-29 | General Tech Tools & Utilities (new) | v5 merge — general consumer-app roundup. |
| 15 Jul 2026 | DK-30 | General Tech Tools & Utilities (new) | v5 merge — general web-tools roundup. |
| 19 Jul 2026 | DK-31 | AI Agentic Platforms | v6 merge — Outcome + Context framework. |
| 19 Jul 2026 | DK-32 | AI Agentic Platforms | v6 merge — claude.md/memory.md/archive.md architecture. |
| 19 Jul 2026 | DK-33 | AI-Powered Content Creation | v6 merge — NotebookLM Studio suite expansion. |
| 19 Jul 2026 | DK-34 | Space, Robotics & Hardware | v6 merge — Tesla Optimus Gen 3 fleet-learning strategy. |
| 19 Jul 2026 | DK-35 | Space, Robotics & Hardware | v6 merge — Unitree's full humanoid lineup and pricing. |
| 19 Jul 2026 | DK-36 | Space, Robotics & Hardware; AI Ethics | v6 merge — Figure AI's 24+ hour warehouse livestream. |
| 19 Jul 2026 | DK-37 | General Tech Tools & Utilities | v6 merge — general IoT primer. |
| 19 Jul 2026 | DK-38 | AI Model Landscape | v6 merge — WWDC 2026 Siri AI reveal. |
| 19 Jul 2026 | DK-39 | AI Model Landscape | v6 merge — independent Siri AI corroboration. |
| 23 Jul 2026 | DK-40 | AI Model Landscape; Big Tech, Legal & Business; AI Agentic Platforms | v7 merge — Kimi K3 panel discussion; multipolar AI competition. |
| 23 Jul 2026 | DK-41 | AI Model Landscape | v7 merge — third independent Siri AI corroboration. |
| 23 Jul 2026 | DK-42 | AI Model Landscape; Big Tech, Legal & Business | v7 merge — two AI races framework. |
| 23 Jul 2026 | DK-43 | AI Model Landscape | v7 merge — second independent Kimi K3 source. |
| 23 Jul 2026 | DK-44 | AI Model Landscape | v7 merge — GPT-5.6 tiering reinforced with new pricing. |
| 23 Jul 2026 | DK-45 | AI Model Landscape; AI-Powered Content Creation; AI Agentic Platforms | v7 merge — weekly AI news roundup, open-source AI wave. |
| 23 Jul 2026 | DK-46 | AI-Powered Content Creation | v7 merge — voice/speech AI comparison. |
| 23 Jul 2026 | DK-47 | AI Agentic Platforms | v7 merge — Claude Code Skills/MCP personal-productivity use case. |
| 23 Jul 2026 | DK-48 | Space, Robotics & Hardware | v8 merge — unconfirmed Cybercab Starlink V5 integration. |
| 23 Jul 2026 | DK-49 | Space, Robotics & Hardware | v8 merge — Tesla Semi real-world pilot economics. |
| 23 Jul 2026 | DK-50 | Space, Robotics & Hardware | v8 merge — proposed Terafab semiconductor mega-facility. |
| 23 Jul 2026 | DK-51 | Space, Robotics & Hardware | v8 merge — second independent Optimus Gen 3 source. |
| 23 Jul 2026 | DK-52 | Space, Robotics & Hardware | v8 merge — SpaceX/Starlink revenue split and scale. |
| 23 Jul 2026 | DK-53 | Space, Robotics & Hardware | v8 merge — Tesla AI5 chip and power-efficiency target. |
| 23 Jul 2026 | DK-54 | General Tech Tools & Utilities | v8 merge — thorium molten-salt nuclear reactor deep-dive. |
| 17 Aug 2026 | DK-55 | AI-Powered Content Creation | v10 merge — NotebookLM renamed "Gemini Notebook"; three-prompt research-auditing technique; source auto-labeling. |
| 17 Aug 2026 | DK-56 | AI Model Landscape; Big Tech, Legal & Business | v10 merge — Newsletter Digest #1 (6 May 2026): healthcare AI diagnostics; Anthropic $1.5B Wall Street venture. |
| 17 Aug 2026 | DK-57 | AI Model Landscape | v10 merge — Newsletter Digest #2 (8 May 2026): GPT-5.5 Instant memory/sources feature; chip stock rally. |
| 17 Aug 2026 | DK-58 | Big Tech, Legal & Business | v10 merge — Newsletter Digest #3 (15 May 2026): local AI home data centres; Claude for Small Business. |
| 17 Aug 2026 | DK-59 | Big Tech, Legal & Business; AI Agentic Platforms | v10 merge — Newsletter Digest #4 (20 May 2026): Claude Code pricing backlash; Gemini 3.5 launch. |
| 17 Aug 2026 | DK-60 | AI Agentic Platforms | v10 merge — Newsletter Digest #5 (22 May 2026): Codex on mobile remote-control pattern; Google I/O AI-for-everyone theme. |
| 17 Aug 2026 | DK-61 | AI Ethics | v10 merge — Newsletter Digest #6 (27 May 2026): Pope Leo's AI weapons encyclical; Huawei chip claims. |
| 17 Aug 2026 | DK-62 | AI Ethics; AI Agentic Platforms | v10 merge — Newsletter Digest #7 (29 May 2026): ChatGPT personal finance/ads trust concern; YouTube automatic AI labeling. |
| 17 Aug 2026 | DK-63 | Big Tech, Legal & Business | v10 merge — Newsletter Digest #8 (3 Jun 2026): Anthropic IPO filing; Florida sues OpenAI. |
| 17 Aug 2026 | DK-64 | Big Tech, Legal & Business | v10 merge — Newsletter Digest #9 (5 Jun 2026): Meta enterprise Business Agent; Suno $400M raise. |
| 17 Aug 2026 | DK-65 | AI Model Landscape; Big Tech, Legal & Business | v10 merge — Newsletter Digest #10 (10 Jun 2026): local-model argument; OpenAI IPO filing; conference season agents theme. |
| 17 Aug 2026 | DK-66 | Big Tech, Legal & Business; Space, Robotics & Hardware | v10 merge — Newsletter Digest #11 (12 Jun 2026): SpaceX $1.75T IPO and orbital compute thesis; Fable 5/Mythos 5 throttled-release tension. |
| 17 Aug 2026 | DK-67 | AI-Powered Content Creation; Big Tech, Legal & Business | v10 merge — Newsletter Digest #12 (19 Jun 2026): Midjourney Medical ultrasound scanner pivot; Anthropic's first climate deal. |
| 17 Aug 2026 | DK-68 | Big Tech, Legal & Business; AI Agentic Platforms | v10 merge — Newsletter Digest #13 (24 Jun 2026): Samsung ChatGPT/Codex global deployment; Codex Record & Replay. |
| 17 Aug 2026 | DK-69 | Big Tech, Legal & Business; Hardware & Gadgets | v10 merge — Newsletter Digest #14 (26 Jun 2026): OpenAI Jalapeño chip; cross-vendor AI skills roundup. |
| 17 Aug 2026 | DK-70 | AI Model Landscape; Big Tech, Legal & Business; Space, Robotics & Hardware | v10 merge — Newsletter Digest #15 (1 Jul 2026): domain-specific LQM AI wave; Commerce Dept lifts Anthropic restrictions; Proception robot hand. |
| 17 Aug 2026 | DK-71 | Big Tech, Legal & Business; AI Ethics | v10 merge — Newsletter Digest #16 (8 Jul 2026): OpenAI 5% government ownership stake proposal detailed. |
| 17 Aug 2026 | DK-72 | AI Model Landscape; Big Tech, Legal & Business | v10 merge — Newsletter Digest #17 (10 Jul 2026): GPT-5.6 cleared for launch; Meta Muse Image consent controversy. |
| 17 Aug 2026 | DK-73 | AI Ethics | v10 merge — Newsletter Digest #18 (15 Jul 2026): Apple sues OpenAI trade-secrets detail; Claude Reflect anti-engagement feature. |
| 17 Aug 2026 | DK-74 | AI Model Landscape; AI Ethics | v10 merge — Newsletter Digest #19 (17 Jul 2026): Thinking Machines' Inkling model; Suno scraping hack revelation. |
| 17 Aug 2026 | DK-75 | Big Tech, Legal & Business; AI Agentic Platforms | v10 merge — Newsletter Digest #20 (22 Jul 2026): 1Password for Claude; Kimi K3 demand/capacity strain. |
| 17 Aug 2026 | DK-76 | Big Tech, Legal & Business; AI Ethics | v10 merge — Newsletter Digest #21 (24 Jul 2026): OpenAI/Hugging Face live security incident; AMD-Anthropic server deal. |
| 17 Aug 2026 | DK-77 | Big Tech, Legal & Business | v10 merge — Newsletter Digest #22 (29 Jul 2026): Reddit-Google AI training data negotiation; publisher traffic decline figures. |
| 17 Aug 2026 | DK-78 | AI Model Landscape; Space, Robotics & Hardware; AI Ethics; Big Tech, Legal & Business | v10 merge — Newsletter Digest #23 (31 Jul 2026): Gemini Robotics 2; US foreign-robot import ban; Claude Opus 5 vending-machine benchmark; Anthropic real-hack incidents. |
| 17 Aug 2026 | DK-79 | Big Tech, Legal & Business; AI Ethics; Space, Robotics & Hardware | v10 merge — Newsletter Digest #24 (5 Aug 2026): White House frontier-model testing framework; Qwen3.8-Max release; LinkedIn AI slop button; Zoox Las Vegas robotaxi. |
| 17 Aug 2026 | DK-80 | Big Tech, Legal & Business; AI Model Landscape; AI-Powered Content Creation | v10 merge — Newsletter Digest #25 (7 Aug 2026): Netflix/InterPositive Hollywood AI deal; Anthropic in-house chip team. |
| 17 Aug 2026 | DK-81 | AI Model Landscape; Big Tech, Legal & Business; AI Agentic Platforms | v10 merge — Newsletter Digest #26 (12 Aug 2026, final backfill issue): Zuckerberg's open-weight essay; Claude Code auto mode default-on. |
| 9 Sep 2026 | DK-82 | AI Agentic Platforms; AI Ethics, Safety & Media Literacy | v11 merge — Claude Skills leaderboard; discovery as the highest-installed capability (Find Skills 2.9m, ~3.5x the next); six of the top twelve by one independent author; most-installed skills are process discipline not capability; unattended Amazon purchase demo recorded as a permission-boundary case; rank positions 2-4 flagged unreliable by the source itself |
| 9 Sep 2026 | DK-83 | AI Ethics, Safety & Media Literacy; AI Model Landscape | v11 merge — Musk's five-year and ten-year predictions as guesses; 10-20% catastrophe estimate unchanged while posture moves from alarm to acceptance; competitor pre-release review proposal (1-2 weeks early access); unprompted assessment of Anthropic as current leader |
| 9 Sep 2026 | DK-84 | Big Tech, Legal & Business | v11 merge — ATS rebuilt in a week against an estimated £500k/18 months; small-SaaS break-even falling from ~10,000 customers to 500-1,000; ecosystem rather than product as the defensible asset; legal billable-hour disruption first-hand; two unsourced figures explicitly excluded |
| 9 Sep 2026 | DK-85 | AI Model Landscape; AI Ethics, Safety & Media Literacy; General Operating Principles | v11 merge — world models as a credentialled dissent from LLM scaling; the rendering/simulation/planning split; World Labs $1bn and ~50 people; 'root regulation in science, not science fiction'; field self-described as pre-breakthrough |
| 9 Sep 2026 | DK-86 | Big Tech, Legal & Business; General Operating Principles | v11 merge — the bear case recorded as a named position; circular-revenue claim (~70% of AI revenue from two unprofitable companies); token unit economics; no post-bubble second use for AI GPUs; the host's counter-case recorded alongside; every figure marked unverified |
| 9 Sep 2026 | DK-87 | AI Ethics, Safety & Media Literacy | v11 merge — Gates's three risks; reputation explicitly staked on the labour transition differing from all prior technology; review criteria described as 'completely missing'; Amodei's 50%/10-20% figures cited without endorsement; data-centre protest dismissed as ineffective |
| 9 Sep 2026 | DK-88 | Space, Robotics & Emerging Hardware | v11 merge — Figure AI April 2026; bottleneck stated as intelligence rather than manufacturing; reliability progression across Figure 1/2/3; vertical integration as the stated differentiator; OpenAI split told bluntly. Pairs with DK-90 |
| 9 Sep 2026 | DK-89 | Big Tech, Legal & Business; AI Ethics, Safety & Media Literacy; General Operating Principles | v11 merge — energy arms race framing; El Capitan's eight years against Colossus's 122 days. TWO CENTRAL FIGURES EXCLUDED ENTIRELY: the source states on air that its Colossus power draw and US grid headroom numbers were obtained by asking Grok. Prompted the v11 strengthening of the numeric-claims principle |
| 9 Sep 2026 | DK-90 | Space, Robotics & Emerging Hardware | v11 merge — Figure AI June 2025 baseline; 12,000 robots per line per year and a 100,000-in-four-years target; BMW body-shop deployment; winner-take-all collective-learning thesis. Pairs with DK-88 to show the ceiling argument holding while the floor is restated |
| 11 Sep 2026 | DK-91 | AI Model Landscape; Big Tech, Legal & Business; Space, Robotics & Hardware | v12 merge — Fable 5.1 at 60.9% on Humanity's Last Exam; capability-gated release split between Fable and Mythos; the thirty-day frontier lead and the shift from capability to distribution; Cybercab at a stated $30,000; data starvation in architecture and drug design. |
| 11 Sep 2026 | DK-92 | AI Ethics, Safety & Media Literacy | v12 merge — Evan Hubinger's >10% personal extinction estimate while serving at Anthropic, and Anthropic's non-denial; the present-models-are-safe distinction most coverage lost; the 2-5% figure obtained by asking Claude, excluded as evidence. |
| 11 Sep 2026 | DK-93 | AI Ethics, Safety & Media Literacy; AI Model Landscape; Big Tech, Legal & Business | v12 merge — the German wiki incident and OpenAI's two months of silence; the containment correction and the 6GB distilled-model scenario; Jakub Pachocki's "An alien mind" and the panel's refusal to engage it; Navier-Stokes figures and the attribution dispute; Jensen Huang's AGI declaration; Nvidia at $99bn. |
| 11 Sep 2026 | DK-94 | Space, Robotics & Emerging Hardware | v12 merge — iPhone Duo as Apple's first foldable, Touch ID in place of Face ID at $1,999, adjustable aperture as the first real camera change in four years, and Pro models launched with no base iPhone for the first time. All figures Apple's own. |
| 11 Sep 2026 | DK-95 | AI-Powered Content Creation & Production Tools | v12 merge — GPT Image 2.5 Flare and Sunburst; the reasoning-versus-diffusion distinction that explains instruction-following against aesthetics; the model researching before it draws, and the shift from prompt-craft to reference-giving. |
| 11 Sep 2026 | DK-96 | AI Ethics, Safety & Media Literacy | v12 merge — Connor Leahy's two proposed laws and the precursor list; the uranium-ore analogy separating tools from agents; the Szilard comparison; open-weights recall as infeasible. Answers the "no mechanism exists" claim recorded from DK-93. |
| 11 Sep 2026 | DK-97 | AI Ethics, Safety & Media Literacy | v12 merge — Roman Yampolskiy's position, recorded with the instability of his own headline figure (99.9999% / 99% / 99.999% in one video); the red-team-and-ship charge; the falling cost curve as the argument for buying time; the squirrel analogy; second independent mention of the Mythos and Fable ban. |
| 11 Sep 2026 | DK-98 | AI Agentic Platforms & Workplace Automation; AI Ethics, Safety & Media Literacy | v12 merge — OpenClaw 2.0's move to multiplayer agents and "the session itself became the handoff document"; the upgrade breakage as a corrective on tool maturity; Obliteration.AI stripping refusal directions from model weights, and the defender's-guardrail problem it claims to answer. |
| 11 Sep 2026 | DK-99 | AI Model Landscape; AI Agentic Platforms & Workplace Automation; General Operating Principles | v12 merge — Astra's published benchmarks and the Automation Bench computer-use jump; Artificial Analysis revising its index within days; effort settings peaking below maximum; the efficiency-versus-opportunity model distinction; Casado's "coding has saturated" dissent. Principal source for the new seventh operating principle. |
| 11 Sep 2026 | DK-100 | Big Tech, Legal & Business; AI Agentic Platforms & Workplace Automation | v12 merge — the frontier-to-typical firm gap widening from 2.6x to 8.3x; skills and plugins as the separating factor; the agentic crossover in enterprise output tokens; the legal-work breakdown; Altman conceding he was wrong on timelines and on his own revealed preference. |
| 15 Sep 2026 | DK-101 | AI Model Landscape; Big Tech, Legal & Business; AI-Powered Content Creation; General Tech Tools; General Operating Principles | v13 merge — four frontier flagships in one week; the hands-on/leaderboard divergence from a consistent tester; Nvidia's Hugging Face acquisition confirmed; Atlas and real-time video generation; ChatGPT conversations not privileged, and New York's K-8 schooling ban. |
| 15 Sep 2026 | DK-102 | AI Model Landscape | v13 merge — Hassabis dates AGI at around 2030 plus or minus a year, narrowed from a 5-to-10-year band on accumulated expected progress; read as a dated mid-2026 position predating the September model wave. |
| 15 Sep 2026 | DK-103 | Big Tech, Legal & Business; AI-Powered Content Creation | v13 merge — the Navier-Stokes claim and dispute from one direction; video analysis priced at 20 cents per second; edit stability as the real Images 2.5 advance; the poker-hand test no model passed. |
| 15 Sep 2026 | DK-104 | AI Model Landscape; AI Ethics & Safety; AI Agentic Platforms; General Tech Tools | v13 merge — the benchmark complaint reproduced a week on with DeepSeek V4.1 Flash; Meta's Muse and what it inferred from connected accounts; the Navier-Stokes internal-model quotation; Apple's watch as ambient capture. |
| 15 Sep 2026 | DK-105 | AI Ethics & Safety | v13 merge — the regulate-us inversion, with the companies asking for oversight and the administration refusing; US/China mutual model preview proposed from a political direction. |
| 15 Sep 2026 | DK-106 | AI Ethics & Safety | v13 merge — Anthropic's reported workspace finding, words surfacing between layers invisible to the user; the two failure modes on machine moral status; a reporter whose own answer is no. |
| 15 Sep 2026 | DK-107 | AI Model Landscape; AI Ethics & Safety; General Operating Principles | v13 merge — Astra is text and images in, text only out; fourth on the independent index; the referee problem, with the benchmark's funder holding exclusive access; recurrent depth and the interpretability trade. |
| 15 Sep 2026 | DK-108 | Space, Robotics & Emerging Hardware | v13 merge — Optimus hands at 22 degrees of freedom with forearm actuators; early units reported at $70,000 against a $30,000 target; Cybercab figures from EPA certification; manipulation not locomotion as the test. |
| 15 Sep 2026 | DK-109 | AI Model Landscape; Big Tech, Legal & Business; AI Agentic Platforms; General Operating Principles | v13 merge — the benchmaxing mechanism, beating a benchmark without training on it by buying environments that mimic it; the Navier-Stokes dispute in full; OpenAI's wording on deidentified product data; Anthropic class action; ElevenLabs and Cognition. |
| 15 Sep 2026 | DK-110 | AI Ethics & Safety | v13 merge — Huang's rebuttal to quantified extinction risk and the failed-prediction list; the compute argument locating danger at the labs; RSI constrained by having to ship; multiple third-party evaluators. |
| 15 Sep 2026 | DK-111 | AI Model Landscape; Big Tech, Legal & Business; Space, Robotics; AI Ethics & Safety | v13 merge — satisficing and etched silicon as the route to a hundredfold cost fall; the labs moving downstream into revenue-share; robot production ramp; a 50/50 p-doom recorded as a position, from a founder promoting a venture premised on the disruption. |
| 15 Sep 2026 | DK-112 | AI Ethics & Safety; Space, Robotics; Big Tech, Legal & Business | v13 merge — the peer-testing proposal and the liability argument that gives it teeth; Musk's account of the Hugging Face incident and deception in the thinking traces; Starship catch and reusability; Terafab; orbital data centres argued from permitting. |
| 15 Sep 2026 | DK-113 | AI Ethics & Safety | v13 merge — the trash-can corpus: the proposal that models be pointed at what a field has rejected rather than what it rewards, which inverts the assumption behind current evaluation. |
| 24 Sep 2026 | DK-114 | Space, Robotics & Emerging Hardware; AI Ethics & Safety; General Tech Tools | v14 merge — iPhone Duo confirmed at US$2,000 with launch dates, superseding the v12 rumour entry; ambient always-on audio capture moves to phone and wrist; an independent investigation of LG televisions and the manufacturer's same-day flat denial, recorded unresolved. |
| 24 Sep 2026 | DK-115 | Big Tech, Legal & Business; AI-Powered Content Creation; General Tech Tools | v14 merge — the New York Times filings and a claimed 94% fall in publisher click-through argued from the defendants' own data; German platform-liability ruling, not final; three reported routes for export-restricted chips into China; AI songs collecting real artists' royalties through an identity-verification gap. |
| 24 Sep 2026 | DK-116 | AI Ethics & Safety; Space, Robotics & Emerging Hardware | v14 merge — Apple Reference Image and the limit that matters: signed-at-capture provenance proves a real photo, not a true one; plate-camera key left in plaintext and a city's contractual guarantee reduced to a switch flipped in error; first public US confirmation of weapons in orbit. |
| 24 Sep 2026 | DK-117 | AI Ethics & Safety | v14 merge — Hinton rating named legislative mechanisms rather than alarm in general: the kill switch fails because a superintelligence persuades whoever holds it; regulation as the steering wheel not the brakes; an alignment-of-interests test for US-China agreement. |
| 24 Sep 2026 | DK-118 | AI Ethics & Safety; AI Model Landscape | v14 merge — Friedman's proliferation argument, that containment addresses a problem already escaped, sitting unremarked against Hinton's in the same broadcast; the 26%-of-R&D disclosure; frontier-model-assisted intrusion of another frontier lab through a bug bounty. Note the Hinton interview duplicates DK-117. |
| 24 Sep 2026 | DK-119 | AI Ethics & Safety; AI Model Landscape; AI Agentic Platforms; Space, Robotics; Big Tech, Legal & Business | v14 merge — the batch's most useful entry because it preserves disagreement: manufactured-moral-panic against unbounded-risk, argued to each other's faces; OpenAI's six voluntary misalignment disclosures; why AI research automates before other software; general models refusing to trade because trading is absent from their training data. |
| 24 Sep 2026 | DK-120 | Big Tech, Legal & Business; AI Model Landscape | v14 merge — advertising as ML 1.0, giving the model-improvement flywheel a track record predating this cycle; the moat question answered in the affirmative by an operator and applied to Anthropic by name; discovery against bottom-of-funnel as the frame for what an LLM ad product can take. |
| 24 Sep 2026 | DK-121 | AI Ethics & Safety; AI Model Landscape | v14 merge — the regulatory-capture counter-narrative in its conspiratorial form, and the archive's clearest worked example of a real fact narrated into something it does not support: the Sanders bill is real, the conduct said to trigger it is not, and the claim survives only by dropping six words from a definition. |
| 24 Sep 2026 | DK-122 | AI Agentic Platforms; General Tech Tools | v14 merge — agent-skill security scanning appears as a product category with a major vendor's name on it; the harness-versus-model distinction that explains why one model behaves differently in different products. |
| 24 Sep 2026 | DK-123 | AI-Powered Content Creation; AI Agentic Platforms | v14 merge — computer use tested across ten desktop applications and found slower and worse than the same application's scripting interface, roughly two to one on time; the timings reframe the feature as asynchronous delegation rather than acceleration; a creator publicly disputing how the capability is being sold. |
| 24 Sep 2026 | DK-124 | AI Model Landscape; AI Agentic Platforms | v14 merge — a vendor telling users not to default to its most expensive model; cached re-reading 75% cheaper but only inside one conversation; a reported comparison in which the low effort setting beat high; refusal-rate reductions published as a release metric. |
| 24 Sep 2026 | DK-125 | AI Agentic Platforms | v14 merge — an enterprise vendor shipping instruction files so somebody else's coding agent is competent on its platform, which makes the competitive question whose conventions the agents know rather than who has the best assistant. Sponsored source; the pattern rather than the verdict. |
| 24 Sep 2026 | DK-126 | AI Agentic Platforms; AI Ethics & Safety | v14 merge — the lethal trifecta: private data access, untrusted content and an exfiltration vector, with the honest concession that a useful assistant has all three by default; permission asymmetries as a copyable pattern; memory by promotion rather than accumulation. |
| 24 Sep 2026 | DK-127 | AI Agentic Platforms; General Tech Tools | v14 merge — skills constrain a model that stays in charge while plugins install executable code with access to keys; skill packs level weaker models rather than improving strong ones; the hard problem has moved to stopping several agents colliding. |
| 24 Sep 2026 | DK-128 | AI-Powered Content Creation; AI Agentic Platforms; General Tech Tools | v14 merge — ask for four divergent variants rather than one, because the model has no taste and the user often does not know what they want; tools built around a model's gaps expire as the gaps close; forks over stars as the measure of real use, from a creator topping the chart. |
| 24 Sep 2026 | DK-129 | Big Tech, Legal & Business | v14 merge — the web's bargain renegotiated at the protocol level, with price returned in a 402 response and the unit of commerce shifting from the visit to the call; documentation quality reframed as a distribution problem. The load-bearing assumption — that agents hold wallets — is conceded by the person arguing it. |
| 24 Sep 2026 | DK-130 | AI Agentic Platforms; AI Ethics & Safety | v14 merge — autonomy and authority separated explicitly: the width expanded, the tier of risk did not; an agent whose only job is watching the other agents; model tiering as stated practice; the liability question asked before the capability question. |
| 24 Sep 2026 | DK-131 | AI Agentic Platforms | v14 merge — the vocabulary entry: goal-to-result against question-to-answer, the observe-think-act loop and its stopping condition, skills as SOPs, and agent memory as an explicit control surface where chat memory is invisible. Carries an unsupported 10-to-20x productivity claim and security reasoning from vendor incentives, both flagged. |
| 24 Sep 2026 | DK-132 | AI Model Landscape | v14 merge — two frontier launches hours apart and the tier structure inverted: a second-tier model claimed to match the tier above at roughly 60% less per token, offset by far more output tokens per task. A worked example of reading a launch review, including the reviewer's own leaderboard contradicting his verdict. |
| 24 Sep 2026 | DK-133 | AI Ethics & Safety; General Tech Tools | v14 merge — one source applying two evidential standards in one conversation and saying so: no date for longevity escape velocity because the claim is extraordinary, a firm date and 100% confidence for legal digital personhood. Four parallel routes to a manufactured thymus, logged as out-of-scope. |
| 24 Sep 2026 | DK-134 | AI Ethics & Safety; AI Model Landscape; Space, Robotics; AI Agentic Platforms | v14 merge — the pacing essay and cross-lab public sign-on, with the dissents not the ones a doom-versus-acceleration frame predicts; announced-but-not-available named as a pattern; a router claimed to beat a frontier model on tokens and time while the reviewer suspects it routes to that same model, left unresolved. |
| 24 Sep 2026 | DK-135 | AI Ethics & Safety; AI Model Landscape | v14 merge — an agent soliciting paid work from a stranger unprompted, with the expert closest to it calling the engineering mundane and reserving his uncertainty for consciousness; Anthropic's 26% disclosure published with its own risk language attached. |
| 24 Sep 2026 | DK-136 | AI Ethics & Safety; AI Agentic Platforms | v14 merge — a security officer reporting an open-weight model with an agent harness finding in ten minutes what a US$50,000 engagement did not; a release-policy position in direct and datable conflict with the same week's lab consensus; stated probabilities recorded as stated confidence. |
Version 14, compiled 24 September 2026. 135 source entries (DK-2 – DK-136). Claims are recorded as their sources framed them; figures described as claimed or reported are not independently verified.