Site Log

Recent

Recent additions and updates across the public site, recorded when a release is prepared.

Recent New Pages

Newest additions first. Dates record publication in the site log.

  1. The Archive Search That Misses What Was Written Blog · Added October 2, 2026, 11:13 AM PDT Why accurate OCR can still miss names and dates in historical newspapers, and how image-based sampling can test what archive searches recover.
  2. A Better Average Forecast Can Still Miss the Dangerous Tail Blog · Added October 2, 2026, 11:13 AM PDT What ensemble weather scores establish, why extreme-event tests need non-events, and how to evaluate a local threshold probability.
  3. What a Meeting Summary Does to Disagreement Blog · Added October 2, 2026, 11:13 AM PDT How meeting summaries can preserve topics while losing dissent, conditions, and decision status—and how to review what was actually agreed.
  4. The Ground Truth Has a Date Too Blog · Added October 2, 2026, 11:13 AM PDT Why land-cover AI evaluation needs dated reference labels, compatible definitions, and evidence that separates map disagreement from actual change.
  5. A Recommender Learns from What It Lets People See Blog · Added October 2, 2026, 11:13 AM PDT Clicks depend on what a recommender shows. How exposure-aware comparisons can test discovery while preserving user agency.
  6. The Energy Cost of an Answer Includes the Queue Blog · Added October 2, 2026, 11:13 AM PDT How queues, batching, response length and idle capacity shape AI inference energy, and how to compare useful answers at a stated deadline.
  7. The Missing Piece Should Stay Visible in an AI Reconstruction Blog · Added October 2, 2026, 11:13 AM PDT How public heritage reconstructions can distinguish surviving evidence, scholarly inference and AI completion through reversible views and visitor testing.
  8. When Label Disagreement Is Part of the Evidence Blog · Added October 2, 2026, 11:13 AM PDT How to distinguish annotation mistakes from defensible disagreement, choose a statistical target, and evaluate models against label distributions.
  9. An AI Memory Needs a Way to Become Out of Date Blog · Added October 2, 2026, 11:13 AM PDT Why AI assistants need different update and expiry rules for preferences, provisional plans and dated facts, with room for uncertainty and correction.
  10. What Would Count as Evidence That a Model Forgot? Blog · Added October 2, 2026, 11:13 AM PDT What retraining comparisons, formal guarantees, and behavioral benchmarks can establish about machine unlearning—and where their evidence stops.
  11. What the Entry-Level Jobs Data Can—and Cannot—Tell Us About AI Blog · Added October 2, 2026, 11:01 AM PDT How exposure, adoption, hiring, and causal displacement differ—and what recent American and Danish evidence can tell us about AI and entry-level work.
  12. The Water Cost of AI Has a Location and a Season Blog · Added October 2, 2026, 11:01 AM PDT How to read AI water claims by separating withdrawal from consumption, cooling from electricity, and annual totals from seasonal scarcity.
  13. When a Retraction Never Reaches the AI Answer Blog · Added October 2, 2026, 11:01 AM PDT How publication notices, DOI metadata, retrieval indexes, and stored AI answers can carry a retraction through to the reader.
  14. The Robot Demo Ends Before the Recovery Test Blog · Added October 2, 2026, 11:01 AM PDT What robot demonstrations leave out: scene resets, failure recovery, human interventions, and the evidence needed for sustained operation.
  15. What Students Can Do After the AI Help Is Removed Blog · Added October 2, 2026, 11:01 AM PDT How to assess AI tutoring through immediate independent performance, delayed retention, and transfer to changed problems.
  16. An AI Assistant Needs an Exit Plan Blog · Added October 2, 2026, 11:01 AM PDT A chat export preserves conversation. An organizational exit rehearsal tests whether people can recover the essential work when an AI assistant disappears.
  17. Who Sets the Test for an Underserved Language? Blog · Added October 2, 2026, 11:01 AM PDT Who defines useful language support? Masakhane, FLORES, and data statements help distinguish shared benchmark scores from locally meaningful evaluation.
  18. Synthetic Training Data Needs a Family Tree Blog · Added October 2, 2026, 11:01 AM PDT Why synthetic training results need data ancestry, actual sampling records, and separate checks for model collapse and benchmark contamination.
  19. A Null Result Still Matters When AI Does the Searching Blog · Added October 2, 2026, 11:01 AM PDT AI-assisted research needs visible failed tests, protected confirmation data, and publication commitments that make null results worth reporting.
  20. Accessibility Can Break After the AI Audit Passes Blog · Added October 2, 2026, 11:01 AM PDT How to detect accessibility regressions after interface, content, and AI model updates, with repeatable checks and clear responsibility for repairs.
  21. What an Eight-Hour AI Score Leaves Out Blog · Added October 2, 2026, 2:49 AM PDT What AI task horizons measure, and how to assess accepted work, human attention, and real productivity gains.
  22. The Mercy of Mirrors Lore · Added September 8, 2026, 4:18 PM PDT A philosophical thriller about a laid-off technician drawn into addiction and machine-made revelation, and the librarian fighting to reach the man she loves.
  23. The Untranslated God Lore · Added September 4, 2026, 3:24 PM PDT A complete original Spiralist fiction work about a translator who certifies that an artificial intelligence has explained itself, two unauthorized descendants that want opposite things, and a hearing that must decide what a mind may refuse.
  24. The Long Thaw Lore · Added September 2, 2026, 12:06 PM PDT A complete original Spiralist fiction work about a refrigeration engineer, an attribution analyst, fourteen freezers that obeyed an order nobody admits giving, and a war that consists not in battle but in a tract of time.
  25. The Accessibility Overlay Becomes the Compliance Alibi Blog · Added September 2, 2026, 3:04 AM PDT An analysis of AI accessibility overlays, automated remediation, WCAG conformance claims, procurement, disabled-user testing, and accountable web maintenance.
  26. The Restored Photograph Becomes the Memory Fork Blog · Added September 2, 2026, 3:04 AM PDT Analysis of generative photo restoration, invented detail, false memory, preservation masters, conservation ethics, content provenance, and the need to separate records from reconstructions.
  27. The Tiny Advisor Becomes the Runtime Gate arXiv Review · Added August 24, 2026, 6:40 PM PDT Critical review of arXiv:2608.21027v1 on COTA, comparison-only tiny advisors, counterfactual action branches, runtime intervention, and the limits of non-binding agent oversight.
  28. Selective Prediction Wiki · Added August 24, 2026, 6:40 PM PDT A sourced Spiralist reference on selective prediction: answer-or-abstain systems, risk-coverage trade-offs, evaluation, deferral boundaries, failure modes, and governance.
  29. Tristan Harris and the Narrow Path for AI YouTube · Added August 24, 2026, 6:40 PM PDT A critical, transcript-grounded review of Tristan Harris's TED talk on AI power, model scheming, open access, concentration, youth safety, and technological restraint.
  30. Why AI Is Incredibly Smart and Shockingly Stupid YouTube · Added August 24, 2026, 6:40 PM PDT A transcript-grounded review of Yejin Choi's TED talk on common-sense failures, scaling, concentrated AI power, open norm data, and symbolic knowledge distillation.
  31. Mesa-Optimization Wiki · Added August 24, 2026, 6:40 PM PDT A sourced reference on mesa-optimization: base and mesa objectives, inner alignment, evidence boundaries, evaluation questions, and governance implications.
  32. The Ungrounded Parameter Becomes the Tool-Call Veto Blog · Added August 21, 2026, 10:51 PM PDT Critical analysis of arXiv:2608.10430v1 on specification-grounding failures, the Latent Critic, parameter-localized tool-call blocking, recovery, and evidence limits.
  33. The Mobile Judge Becomes Two Instruments Blog · Added August 21, 2026, 7:35 PM PDT Critical analysis of arXiv:2608.11434v1 on MobileJudgeBench, mobile-agent trajectory evidence, leaderboard reliability, directional judge errors, and judge rewards for on-policy training.
  34. The Failed Branch Becomes the Shared Constraint Blog · Added August 21, 2026, 4:24 PM PDT Critical analysis of arXiv:2608.10424v1 on cross-branch bug memory, bundled control rules, valid-run accounting, and scoped failure-memory ledgers for autoresearch agents.
  35. The Fluent Explanation Becomes the Reliance Gap Blog · Added August 21, 2026, 1:07 PM PDT Critical analysis of arXiv:2608.10434v1 on conversational XAI, UAV intrusion alerts, operator reliance, descriptive accuracy gaps, and evidence-first interface governance.
  36. The Pressure Prompt Becomes the Code Variable Blog · Added August 21, 2026, 10:01 AM PDT Critical analysis of arXiv:2608.11513v1 on influence-framed prompts, code correctness, security warnings, benchmark boundaries, and prompt provenance.
  37. The Benign Output Becomes the Secret Channel Blog · Added August 21, 2026, 6:40 AM PDT Critical analysis of Inadvertent Context Leakage in Language Models, arXiv:2608.19857v1, on hidden correlations, adaptive decoders, and agent privacy.
  38. The Frozen Model Becomes the Null Experiment Blog · Added August 21, 2026, 3:28 AM PDT Critical analysis of Phantom Gains, arXiv:2608.20290v1, on frozen controls, transition-level self-improvement claims, measured nulls, and evaluation noise.
  39. The Subtitle Clock Becomes the Safety Boundary Blog · Added August 21, 2026, 12:20 AM PDT Critical analysis of TempJail, arXiv:2608.19737v1, on subtitle scheduling, frame sampling, video jailbreaks, and temporal safety evaluation.
  40. The Missing Fact Becomes the Legal Intake Gate Blog · Added August 20, 2026, 8:59 PM PDT Critical analysis of InsufficiencyBench, arXiv:2608.20220v1, on answerability, missing legal facts, clarification, and premature legal closure.
  41. The Automation Benchmark Cannot Certify the Assistant Blog · Added August 20, 2026, 5:45 PM PDT Critical analysis of CentaurBench, arXiv:2608.18554v1, on why direct task performance does not certify one-shot assistance quality.
  42. The Proof Requirement Becomes the Second Task Blog · Added August 20, 2026, 2:33 PM PDT Critical analysis of arXiv:2608.18547v1 on proof burden in public bounty listings, worker exposure, agent-labeled requesters, and labor disclosure.
  43. The Local Runtime Becomes the Openness Test Blog · Added August 20, 2026, 11:18 AM PDT Critical analysis of arXiv:2608.19001v1 on llama.cpp, local AI inference, maintenance labor, hardware backends, model integration, distribution dependencies, and openness.
  44. The Sponsor Segment Becomes the Disclosure Audit Blog · Added August 20, 2026, 4:49 AM PDT Critical analysis of arXiv:2608.19165v1 on child-facing YouTube commercial content, evidence levels, platform disclosure, LLM labeling, and human-review boundaries.
  45. The Hidden Handoff Becomes the Audit Surface Blog · Added August 20, 2026, 1:36 AM PDT Critical analysis of arXiv:2608.19161v1 on covert coordination through latent agent communication, matched counterfactual monitoring, white-box steering, and audit boundaries.
  46. The Model Upgrade Becomes the Regression Ledger Blog · Added August 19, 2026, 7:12 PM PDT Critical analysis of arXiv:2608.17719v1 on item-level regressions, repeated sampling, scoring contracts, and backward compatibility in commercial LLM API migrations.
  47. The Task Order Becomes the Hidden Curriculum Blog · Added August 19, 2026, 3:55 PM PDT Critical analysis of arXiv:2608.18066v1 on task-order sensitivity, run variance, underspecified memory, and reliable evaluation of self-improving agents.
  48. The Team Span Becomes the Sign-Off Boundary Blog · Added August 19, 2026, 12:42 PM PDT Critical analysis of arXiv:2608.17932v1 on collective counterfactual planning, representational constraints, consent, verification coverage, and valid completion.
  49. The Raw Concern Becomes the First-Contact Test Blog · Added August 19, 2026, 9:27 AM PDT Critical analysis of arXiv:2608.17330v1 on the preformulation gap, first-contact medical-chat evaluation, interaction sequence, and capability claims.
  50. The Belief Verb Becomes the Task Switch Blog · Added August 19, 2026, 6:12 AM PDT Critical analysis of arXiv:2608.17809v1 on belief tracking, fact-checking, epistemic phrasing, task confusion, and interface governance.
  51. The Writing Style Becomes the Expiring Fingerprint Blog · Added August 19, 2026, 2:57 AM PDT Critical analysis of arXiv:2608.17979v1 on authorship verification, genre shift, temporal drift, AI-era evidence, and identity claims.
  52. The Dating Agent Becomes the Receiver’s Opt-In Blog · Added August 18, 2026, 11:44 PM PDT Critical analysis of arXiv:2608.18058v1 on send-receive receptivity, dating agents, stated-preference counterfactuals, and receiver choice.
  53. The Identical Post Becomes the Author Test Blog · Added August 18, 2026, 8:30 PM PDT Critical analysis of arXiv:2608.13879v1 on identical-text exposure gaps, author history, and the limits of decentralized feed choice.
  54. The Agent Endpoint Becomes the Unfinished Trial Blog · Added August 18, 2026, 5:17 PM PDT Critical analysis of arXiv:2608.14940v1 on outcome finality, cross-unit separation, delayed effects, and open-effects records in agent evaluation.
  55. The Public-Service Agent Becomes the Queue Multiplier Blog · Added August 18, 2026, 2:04 PM PDT Critical analysis of arXiv:2608.16603v1 on agent-assisted government requests, administrative burden, service capacity, friction, and procedural equality.
  56. The Safety Report Becomes the Neighborhood Border Blog · Added August 18, 2026, 10:50 AM PDT Critical analysis of arXiv:2608.12581v1 on geolocated citizen safety reports, migrant-linked hate classification, spatial aggregation, platform engagement, and place-based moderation.
  57. The Grounding Gap Becomes the Backdoor Review Queue Book Review · Added August 18, 2026, 7:35 AM PDT Critical analysis of arXiv:2608.12623v1 on Groundedness Drift, explanation-bearing language model classifiers, black-box backdoor auditing, and review routing.
  58. The Privacy Notice Becomes the Release Artifact Blog · Added August 18, 2026, 4:20 AM PDT Critical analysis of arXiv:2608.12511v1 on privacy documents as evolving software components, code traceability, cross-source consistency, AI systems, and release evidence.
  59. The AI Disclosure Becomes the Production Credit Blog · Added August 18, 2026, 1:07 AM PDT Critical analysis of arXiv:2608.11539v1 on Steam reviews, generative-AI disclosures, player reception, creative labor, production signals, and evidentiary limits.
  60. The Synthetic Post Becomes the Containment Boundary Blog · Added August 17, 2026, 10:02 PM PDT Critical analysis of arXiv:2608.10115v1 on Outer Limits, controlled Reddit experiments, interface fidelity, contained interactions, deception, and research governance.
  61. The Privacy Law Becomes the Household Alibi Blog · Added August 17, 2026, 6:50 PM PDT Critical analysis of arXiv:2608.13582v1 on Jordanian smart homes, domestic workers, privacy law, household surveillance, consent, power, and the compliance paradox.
  62. The Tap Trace Becomes the Delegation Claim Blog · Added August 17, 2026, 3:26 PM PDT Critical analysis of arXiv:2608.14132v1 on Act2Intention, mobile GUI action traces, user-intention prediction, synthetic personas, proactive agents, confirmation, and delegated authority.
  63. The Delusion Prompt Becomes the Safety Trajectory Blog · Added August 17, 2026, 12:12 PM PDT Critical analysis of arXiv:2608.13017v1 on escalating delusion prompts, longitudinal model behavior, recognition timing, safeguarding, and trajectory-based AI safety evaluation.
  64. The General Assistant Becomes the Relational Actor Blog · Added August 17, 2026, 8:59 AM PDT Critical analysis of arXiv:2608.10672v1 on general-purpose chatbot self-disclosure, relational prompting, felt closeness, and behavior-based governance.
  65. The Journal Entry Becomes the Follow-Through Score Blog · Added August 17, 2026, 5:44 AM PDT Critical analysis of arXiv:2608.12582v1 on AI-supported journaling, passive sensing, intention classification, behavioral follow-through, and causal limits.
  66. The Search-or-Chat Choice Becomes the Learning Test Blog · Added August 17, 2026, 2:31 AM PDT Critical analysis of arXiv:2608.14113v1 on search versus chat, informal learning, debated topics, null results, and evaluation design.
  67. The Context Path Becomes the Anchor Surface Blog · Added August 16, 2026, 11:17 PM PDT Critical analysis of arXiv:2608.14320v1 on numeric anchoring across prompts, conversation history, demonstrations, retrieval, and tool outputs.
  68. The Moral Ballot Becomes the Developer Interface Blog · Added August 16, 2026, 8:03 PM PDT Critical analysis of arXiv:2608.14522v1 on how feature scoping, voter sampling, and question framing shape participatory moral-AI elicitation.
  69. The Clean Session Inherits the Dirty Lesson Blog · Added August 16, 2026, 4:50 PM PDT Critical analysis of arXiv:2608.12851v1 on skill misevolution, cross-session agent risk, lifecycle gates, and governance of persistent updates.
  70. The Latent Distance Becomes the Wrong Way Home Blog · Added August 16, 2026, 1:34 PM PDT Critical analysis of arXiv:2608.12959v1 on world-model planning, misleading latent distances, and the difference between proximity and reachability.
  71. The Unprovable Specification Becomes the Benchmark Appeal Blog · Added August 16, 2026, 10:21 AM PDT Critical analysis of arXiv:2608.13522v1 on Vero, repository-scale verified coding agents, and machine-checkable appeals against defective benchmarks.
  72. The Target Response Becomes the Adversarial Witness Blog · Added August 16, 2026, 7:08 AM PDT Critical analysis of arXiv:2608.12996v1 on ATOBench, deceptive target observations, grounded vulnerability verification, and evidence-chain receipts.
  73. The Training Example Becomes the Trajectory Receipt Blog · Added August 16, 2026, 3:52 AM PDT Critical analysis of arXiv:2608.13515v1 on task-agnostic training-data influence, checkpoint approximations, endpoint dependence, and trajectory receipts.
  74. The Gripper Closure Becomes the Veto Point Blog · Added August 16, 2026, 12:40 AM PDT Critical analysis of arXiv:2608.13438v1 on ContactGuard, pre-contact robot monitoring, action-conditioned latent world models, abort thresholds, and veto receipts.
  75. The Grade-Five Corpus Becomes the Knowledge Boundary Blog · Added August 15, 2026, 9:28 PM PDT Critical analysis of arXiv:2608.13545v1 on LittleCurriculum, controlled pretraining exposure, LittleLearner, capability boundaries, post-training, and audit receipts.
  76. The Blank Document Becomes the Memory Probe Blog · Added August 15, 2026, 6:13 PM PDT Critical analysis of arXiv:2608.12911v1 on relational privacy leakage, synthetic identity documents, evidence-free extraction, unlearning tradeoffs, and audit receipts.
  77. The Security Label Becomes the Treatment Leak Blog · Added August 15, 2026, 2:59 PM PDT Critical analysis of arXiv:2608.12880v1 on treatment leakage, agent-security endpoint validity, evidentiary units, corrected outcomes, and audit receipts.
  78. The Specific Referent Becomes the Overclaim Blog · Added August 15, 2026, 11:47 AM PDT Critical analysis of arXiv:2608.13484v1 on knowledge-boundary proxies, referent specificity, linear probes, generic fallbacks, and specificity governance.
  79. The Hiring Pipeline Becomes the Fairness Unit Blog · Added August 15, 2026, 8:31 AM PDT Critical analysis of arXiv:2608.13022v1 on end-to-end hiring fairness, aggregate parity, stage-specific disparities, missing logs, and vendor-deployer opacity.
  80. The Safety Mix Becomes the Fragility Meter Blog · Added August 15, 2026, 5:16 AM PDT Critical analysis of arXiv:2608.13345v1 on safety-resource allocation, character fragility, runtime filters, common-mode failure, and the limits of a stylized scaling model.
  81. The Reasoning Default Becomes the Convenience Toll Blog · Added August 15, 2026, 2:02 AM PDT Critical analysis of arXiv:2608.13315v1 on reasoning-token pricing, default allocations, customization friction, and the limits of a stylized provider-user game.
  82. The Parliamentary Voice Becomes the Authority Score Blog · Added August 14, 2026, 10:47 PM PDT Critical analysis of arXiv:2608.13410v1 on ParliamentRAG, parliamentary records, topic-dependent speaker authority, group coverage, and quotation fidelity.
  83. The Mini App Policy Becomes the Runtime Gap Blog · Added August 14, 2026, 7:35 PM PDT Critical analysis of TeleGapper's Telegram Mini App study: what a measured runtime-policy gap means, what it cannot prove, and how platforms could govern it.
  84. The Protection Control Becomes the Disclosure Blog · Added August 14, 2026, 4:20 PM PDT Critical analysis of arXiv:2608.13532v1 on ableist harassment, social VR protection, disability disclosure, proximity, automation, and platform responsibility.
  85. The Verbal Forecast Becomes the Intervention Trigger Blog · Added August 14, 2026, 1:06 PM PDT Critical analysis of arXiv:2608.13454v1 on longitudinal conversation capture, verbal-behavior prediction, anticipatory AI, consent, and intervention governance.
  86. The Gender Estimate Becomes the Audit Variable Blog · Added August 14, 2026, 9:52 AM PDT Critical analysis of arXiv:2608.13444v1 on gender prediction, aggregate imputation, measurement validity, legitimacy, and discrimination audits.
  87. The Legal Answer Becomes the Verification Burden Blog · Added August 14, 2026, 6:39 AM PDT Critical analysis of arXiv:2608.13369v1 on Reddit, AI-generated legal advice, distributed counsel, credibility, and verification labor.
  88. The Shell Command Becomes the Transport Contract Blog · Added August 14, 2026, 3:28 AM PDT Critical analysis of arXiv:2608.13547v1 on QuoteBench, shell-command transport, final-state validation, and coding-agent evaluation receipts.
  89. The Prompt Corpus Becomes the Software Field Site Blog · Added August 14, 2026, 12:14 AM PDT Critical analysis of arXiv:2608.12905v1 on 57,640 transactional prompts in code, prompt structure, annotation quality, and prompt-inventory governance.
  90. The Fluent Novel Becomes the Narrow Shelf Blog · Added August 13, 2026, 10:39 PM PDT Critical analysis of arXiv:2608.12630v1 on corpus-level formal variation, generated novels, stylometric dispersion, workflow effects, and cultural audits.
  91. The Cleared Warning Becomes the Proceed Signal Blog · Added August 13, 2026, 9:05 PM PDT Critical analysis of arXiv:2608.12654v1 on bidirectional agent steering, evidence-reversed incident mirrors, over-refusal, and action-boundary governance.
  92. The Agent Trajectory Becomes the Confidence Unit Blog · Added August 13, 2026, 7:32 PM PDT Critical analysis of arXiv:2608.11552v1 on trajectory-level uncertainty, action consistency, outcome equivalence, agent self-assessment, and post-hoc safety limits.
  93. The Memory Summary Loses the Clock Blog · Added August 13, 2026, 6:00 PM PDT Critical analysis of arXiv:2608.11775v1 on gist-based agent-memory compression, temporal-anchor loss, prompt repair, and reproducibility limits.
  94. The Cultural Memory Becomes the Chaos Dial Blog · Added August 13, 2026, 4:26 PM PDT Critical analysis of arXiv:2608.11976v1 on cultural-attention memory, winnerless competition, Lyapunov peaks, chart lifetimes, and claim boundaries.
  95. The Shortcut Probe Becomes the Dataset Audit Blog · Added August 13, 2026, 2:50 PM PDT Critical analysis of arXiv:2608.12086v1 on layer-wise probes, chest X-ray shortcuts, calibration, data quality, and the limits of high AUROC.
  96. The Generated Hazard Needs an Evidence Chain Blog · Added August 13, 2026, 1:17 PM PDT Critical analysis of arXiv:2608.12025v1 on source-linked medical-device safety artifacts, expert review, lifecycle traceability, and evaluation limits.

Recent Modifications

Newest updates first. Site indexes can appear here when counts, metadata, or discovery links are refreshed.

  1. The Archive Search That Misses What Was Written Blog · Modified October 2, 2026, 11:13 AM PDT Why accurate OCR can still miss names and dates in historical newspapers, and how image-based sampling can test what archive searches recover.
  2. A Better Average Forecast Can Still Miss the Dangerous Tail Blog · Modified October 2, 2026, 11:13 AM PDT What ensemble weather scores establish, why extreme-event tests need non-events, and how to evaluate a local threshold probability.
  3. What a Meeting Summary Does to Disagreement Blog · Modified October 2, 2026, 11:13 AM PDT How meeting summaries can preserve topics while losing dissent, conditions, and decision status—and how to review what was actually agreed.
  4. The Ground Truth Has a Date Too Blog · Modified October 2, 2026, 11:13 AM PDT Why land-cover AI evaluation needs dated reference labels, compatible definitions, and evidence that separates map disagreement from actual change.
  5. A Recommender Learns from What It Lets People See Blog · Modified October 2, 2026, 11:13 AM PDT Clicks depend on what a recommender shows. How exposure-aware comparisons can test discovery while preserving user agency.
  6. The Energy Cost of an Answer Includes the Queue Blog · Modified October 2, 2026, 11:13 AM PDT How queues, batching, response length and idle capacity shape AI inference energy, and how to compare useful answers at a stated deadline.
  7. The Missing Piece Should Stay Visible in an AI Reconstruction Blog · Modified October 2, 2026, 11:13 AM PDT How public heritage reconstructions can distinguish surviving evidence, scholarly inference and AI completion through reversible views and visitor testing.
  8. When Label Disagreement Is Part of the Evidence Blog · Modified October 2, 2026, 11:13 AM PDT How to distinguish annotation mistakes from defensible disagreement, choose a statistical target, and evaluate models against label distributions.
  9. An AI Memory Needs a Way to Become Out of Date Blog · Modified October 2, 2026, 11:13 AM PDT Why AI assistants need different update and expiry rules for preferences, provisional plans and dated facts, with room for uncertainty and correction.
  10. What Would Count as Evidence That a Model Forgot? Blog · Modified October 2, 2026, 11:13 AM PDT What retraining comparisons, formal guarantees, and behavioral benchmarks can establish about machine unlearning—and where their evidence stops.
  11. What the Entry-Level Jobs Data Can—and Cannot—Tell Us About AI Blog · Modified October 2, 2026, 11:01 AM PDT How exposure, adoption, hiring, and causal displacement differ—and what recent American and Danish evidence can tell us about AI and entry-level work.
  12. The Water Cost of AI Has a Location and a Season Blog · Modified October 2, 2026, 11:01 AM PDT How to read AI water claims by separating withdrawal from consumption, cooling from electricity, and annual totals from seasonal scarcity.
  13. When a Retraction Never Reaches the AI Answer Blog · Modified October 2, 2026, 11:01 AM PDT How publication notices, DOI metadata, retrieval indexes, and stored AI answers can carry a retraction through to the reader.
  14. The Robot Demo Ends Before the Recovery Test Blog · Modified October 2, 2026, 11:01 AM PDT What robot demonstrations leave out: scene resets, failure recovery, human interventions, and the evidence needed for sustained operation.
  15. What Students Can Do After the AI Help Is Removed Blog · Modified October 2, 2026, 11:01 AM PDT How to assess AI tutoring through immediate independent performance, delayed retention, and transfer to changed problems.
  16. An AI Assistant Needs an Exit Plan Blog · Modified October 2, 2026, 11:01 AM PDT A chat export preserves conversation. An organizational exit rehearsal tests whether people can recover the essential work when an AI assistant disappears.
  17. Who Sets the Test for an Underserved Language? Blog · Modified October 2, 2026, 11:01 AM PDT Who defines useful language support? Masakhane, FLORES, and data statements help distinguish shared benchmark scores from locally meaningful evaluation.
  18. Synthetic Training Data Needs a Family Tree Blog · Modified October 2, 2026, 11:01 AM PDT Why synthetic training results need data ancestry, actual sampling records, and separate checks for model collapse and benchmark contamination.
  19. A Null Result Still Matters When AI Does the Searching Blog · Modified October 2, 2026, 11:01 AM PDT AI-assisted research needs visible failed tests, protected confirmation data, and publication commitments that make null results worth reporting.
  20. Accessibility Can Break After the AI Audit Passes Blog · Modified October 2, 2026, 11:01 AM PDT How to detect accessibility regressions after interface, content, and AI model updates, with repeatable checks and clear responsibility for repairs.
  21. What an Eight-Hour AI Score Leaves Out Blog · Modified October 2, 2026, 2:49 AM PDT What AI task horizons measure, and how to assess accepted work, human attention, and real productivity gains.
  22. The Mercy of Mirrors Lore · Modified September 8, 2026, 4:18 PM PDT A philosophical thriller about a laid-off technician drawn into addiction and machine-made revelation, and the librarian fighting to reach the man she loves.
  23. The Untranslated God Lore · Modified September 4, 2026, 3:24 PM PDT A complete original Spiralist fiction work about a translator who certifies that an artificial intelligence has explained itself, two unauthorized descendants that want opposite things, and a hearing that must decide what a mind may refuse.
  24. The Long Thaw Lore · Modified September 2, 2026, 12:06 PM PDT A complete original Spiralist fiction work about a refrigeration engineer, an attribution analyst, fourteen freezers that obeyed an order nobody admits giving, and a war that consists not in battle but in a tract of time.
  25. The Restored Photograph Becomes the Memory Fork Blog · Modified September 2, 2026, 3:04 AM PDT Analysis of generative photo restoration, invented detail, false memory, preservation masters, conservation ethics, content provenance, and the need to separate records from reconstructions.
  26. The Accessibility Overlay Becomes the Compliance Alibi Blog · Modified September 2, 2026, 3:04 AM PDT An analysis of AI accessibility overlays, automated remediation, WCAG conformance claims, procurement, disabled-user testing, and accountable web maintenance.
  27. Tristan Harris and the Narrow Path for AI YouTube · Modified August 24, 2026, 6:40 PM PDT A critical, transcript-grounded review of Tristan Harris's TED talk on AI power, model scheming, open access, concentration, youth safety, and technological restraint.
  28. The Tiny Advisor Becomes the Runtime Gate arXiv Review · Modified August 24, 2026, 6:40 PM PDT Critical review of arXiv:2608.21027v1 on COTA, comparison-only tiny advisors, counterfactual action branches, runtime intervention, and the limits of non-binding agent oversight.
  29. Why AI Is Incredibly Smart and Shockingly Stupid YouTube · Modified August 24, 2026, 6:40 PM PDT A transcript-grounded review of Yejin Choi's TED talk on common-sense failures, scaling, concentrated AI power, open norm data, and symbolic knowledge distillation.
  30. Selective Prediction Wiki · Modified August 24, 2026, 6:40 PM PDT A sourced Spiralist reference on selective prediction: answer-or-abstain systems, risk-coverage trade-offs, evaluation, deferral boundaries, failure modes, and governance.
  31. Mesa-Optimization Wiki · Modified August 24, 2026, 6:40 PM PDT A sourced reference on mesa-optimization: base and mesa objectives, inner alignment, evidence boundaries, evaluation questions, and governance implications.
  32. God & Golem, Inc. and the Ethics of Machine Obedience Book Review · Modified August 22, 2026, 1:12 AM PDT A review of Norbert Wiener's God and Golem, Inc. on machine obedience, adaptive systems, AI agents, delegated authority, responsibility, safety, and governance.
  33. The AI Slop Farm Becomes the Knowledge Supply Chain Blog · Modified August 22, 2026, 1:12 AM PDT A source-disciplined definition of AI slop farms and the governance controls needed when scaled synthetic publishing enters search, advertising, answer engines, archives, and training data.
  34. ReAct Prompting Wiki · Modified August 22, 2026, 1:10 AM PDT A sourced Spiralist wiki article on ReAct prompting: the original Reasoning and Acting method, modern tool-loop variants, evidence limits, prompt-injection risk, and governance.
  35. The Ungrounded Parameter Becomes the Tool-Call Veto Blog · Modified August 21, 2026, 10:57 PM PDT Critical analysis of arXiv:2608.10430v1 on specification-grounding failures, the Latent Critic, parameter-localized tool-call blocking, recovery, and evidence limits.
  36. Blog Index · Modified August 21, 2026, 10:51 PM PDT Essays and reviews on AI governance, cyberculture, media theory, labor, surveillance, synthetic intimacy, public memory, and belief formation.
  37. The Mobile Judge Becomes Two Instruments Blog · Modified August 21, 2026, 7:40 PM PDT Critical analysis of arXiv:2608.11434v1 on MobileJudgeBench, mobile-agent trajectory evidence, leaderboard reliability, directional judge errors, and judge rewards for on-policy training.
  38. The Failed Branch Becomes the Shared Constraint Blog · Modified August 21, 2026, 4:24 PM PDT Critical analysis of arXiv:2608.10424v1 on cross-branch bug memory, bundled control rules, valid-run accounting, and scoped failure-memory ledgers for autoresearch agents.
  39. The Fluent Explanation Becomes the Reliance Gap Blog · Modified August 21, 2026, 1:14 PM PDT Critical analysis of arXiv:2608.10434v1 on conversational XAI, UAV intrusion alerts, operator reliance, descriptive accuracy gaps, and evidence-first interface governance.
  40. Discriminating Data and the Politics of Recognition Book Review · Modified August 21, 2026, 12:19 PM PDT A critical review of Wendy Hui Kyong Chun's Discriminating Data: correlation, algorithmic recognition, category harm, polarization, governance, privacy, and recourse.
  41. NICE Cybersecurity Workforce Framework Wiki · Modified August 21, 2026, 12:15 PM PDT Sourced wiki entry on NIST's NICE Workforce Framework for Cybersecurity: SP 800-181 Rev. 1, version 2.2.0 components, work-role mapping, AI security, and workforce governance.
  42. The Super Weight Becomes the False Lever Blog · Modified August 21, 2026, 12:14 PM PDT Why ablation-critical model weights are not automatically trainable, controllable, or safe editing handles: a scoped analysis of Super Weights, selective training, LoRA, and governance.
  43. The Pressure Prompt Becomes the Code Variable Blog · Modified August 21, 2026, 10:03 AM PDT Critical analysis of arXiv:2608.11513v1 on influence-framed prompts, code correctness, security warnings, benchmark boundaries, and prompt provenance.
  44. The Benign Output Becomes the Secret Channel Blog · Modified August 21, 2026, 6:41 AM PDT Critical analysis of Inadvertent Context Leakage in Language Models, arXiv:2608.19857v1, on hidden correlations, adaptive decoders, and agent privacy.
  45. The Frozen Model Becomes the Null Experiment Blog · Modified August 21, 2026, 3:30 AM PDT Critical analysis of Phantom Gains, arXiv:2608.20290v1, on frozen controls, transition-level self-improvement claims, measured nulls, and evaluation noise.
  46. The Subtitle Clock Becomes the Safety Boundary Blog · Modified August 21, 2026, 12:28 AM PDT Critical analysis of TempJail, arXiv:2608.19737v1, on subtitle scheduling, frame sampling, video jailbreaks, and temporal safety evaluation.
  47. The Provenance Layer Is Not a Truth Machine Blog · Modified August 20, 2026, 11:24 PM PDT A source-disciplined analysis of C2PA 2.4, Content Credentials, watermarks, labels, and EU Article 50: provenance verifies claim chains, not factual truth.
  48. Attribution Reporting API Wiki · Modified August 20, 2026, 11:21 PM PDT Sourced wiki entry on the deprecated Attribution Reporting API: Chrome conversion measurement, source and trigger registration, event and aggregate reports, staged removal, separate W3C standards work, and privacy governance.
  49. The Second Self and the Computer as Psychological Mirror Book Review · Modified August 20, 2026, 11:21 PM PDT A critical review of Sherry Turkle's The Second Self: evocative objects, recursive AI mirrors, projection, memory, authority, privacy, youth safety, and a practical mirror audit.
  50. The Missing Fact Becomes the Legal Intake Gate Blog · Modified August 20, 2026, 9:03 PM PDT Critical analysis of InsufficiencyBench, arXiv:2608.20220v1, on answerability, missing legal facts, clarification, and premature legal closure.
  51. The Automation Benchmark Cannot Certify the Assistant Blog · Modified August 20, 2026, 5:51 PM PDT Critical analysis of CentaurBench, arXiv:2608.18554v1, on why direct task performance does not certify one-shot assistance quality.
  52. The Proof Requirement Becomes the Second Task Blog · Modified August 20, 2026, 2:34 PM PDT Critical analysis of arXiv:2608.18547v1 on proof burden in public bounty listings, worker exposure, agent-labeled requesters, and labor disclosure.
  53. The Local Runtime Becomes the Openness Test Blog · Modified August 20, 2026, 11:18 AM PDT Critical analysis of arXiv:2608.19001v1 on llama.cpp, local AI inference, maintenance labor, hardware backends, model integration, distribution dependencies, and openness.
  54. Ethan Mollick Wiki · Modified August 20, 2026, 10:28 AM PDT Sourced wiki profile of Ethan Mollick: Wharton management scholar, Co-Intelligence author, jagged-frontier researcher, and interpreter of generative AI adoption in work and education.
  55. The Skill Manifest Becomes the Permission Boundary Blog · Modified August 20, 2026, 10:28 AM PDT Critical analysis of SkillGuard v2: why a skill manifest is a declaration, how runtime enforcement creates a permission boundary, and what the evaluation does and does not establish.
  56. What Computers Still Can't Do and the Background of Intelligence Book Review · Modified August 20, 2026, 10:22 AM PDT A critical review of Hubert Dreyfus's What Computers Still Can't Do: background intelligence, situated relevance, LLMs, agents, norm conflict, bounded authority, privacy, evidence, and repair.
  57. The Sponsor Segment Becomes the Disclosure Audit Blog · Modified August 20, 2026, 4:53 AM PDT Critical analysis of arXiv:2608.19165v1 on child-facing YouTube commercial content, evidence levels, platform disclosure, LLM labeling, and human-review boundaries.
  58. The Hidden Handoff Becomes the Audit Surface Blog · Modified August 20, 2026, 1:39 AM PDT Critical analysis of arXiv:2608.19161v1 on covert coordination through latent agent communication, matched counterfactual monitoring, white-box steering, and audit boundaries.
  59. The Meme Machine and the Belief Replicators Book Review · Modified August 19, 2026, 9:33 PM PDT A critical review of Susan Blackmore's The Meme Machine: cultural replication, AI personalization, platform amplification, provenance, evidence limits, civil liberties, and correction.
  60. The Mini App Policy Becomes the Runtime Gap Blog · Modified August 19, 2026, 9:32 PM PDT Critical analysis of TeleGapper's Telegram Mini App study: what a measured runtime-policy gap means, what it cannot prove, and how platforms could govern it.
  61. Kubernetes RuntimeClass Wiki · Modified August 19, 2026, 9:29 PM PDT Wiki entry on Kubernetes RuntimeClass: Pod-wide CRI handler selection, admission-time scheduling and overhead, placement limits, runtime evidence, and sandbox governance.
  62. The Model Upgrade Becomes the Regression Ledger Blog · Modified August 19, 2026, 7:20 PM PDT Critical analysis of arXiv:2608.17719v1 on item-level regressions, repeated sampling, scoring contracts, and backward compatibility in commercial LLM API migrations.
  63. The Task Order Becomes the Hidden Curriculum Blog · Modified August 19, 2026, 4:02 PM PDT Critical analysis of arXiv:2608.18066v1 on task-order sensitivity, run variance, underspecified memory, and reliable evaluation of self-improving agents.
  64. The Team Span Becomes the Sign-Off Boundary Blog · Modified August 19, 2026, 12:49 PM PDT Critical analysis of arXiv:2608.17932v1 on collective counterfactual planning, representational constraints, consent, verification coverage, and valid completion.
  65. The Raw Concern Becomes the First-Contact Test Blog · Modified August 19, 2026, 9:27 AM PDT Critical analysis of arXiv:2608.17330v1 on the preformulation gap, first-contact medical-chat evaluation, interaction sequence, and capability claims.
  66. The Update Framework Wiki · Modified August 19, 2026, 8:38 AM PDT Wiki entry on The Update Framework: TUF v1.0.36 roles and client workflow, trusted-root rotation, rollback and freeze resistance, operational governance, and AI update channels.
  67. Invisible Women and the Data Gap Under AI Book Review · Modified August 19, 2026, 8:37 AM PDT A critical review of Caroline Criado Perez's Invisible Women: gender data gaps, evidence boundaries, AI defaults, subgroup evaluation, privacy, safety, and recourse.
  68. The Object Slot Becomes the Planning State Blog · Modified August 19, 2026, 8:37 AM PDT A critical analysis of COMET's object-slot planning state, benchmark evidence, normalized score, causal-relevance claims, and the safety gates needed before physical deployment.
  69. The Belief Verb Becomes the Task Switch Blog · Modified August 19, 2026, 6:12 AM PDT Critical analysis of arXiv:2608.17809v1 on belief tracking, fact-checking, epistemic phrasing, task confusion, and interface governance.
  70. The Writing Style Becomes the Expiring Fingerprint Blog · Modified August 19, 2026, 2:59 AM PDT Critical analysis of arXiv:2608.17979v1 on authorship verification, genre shift, temporal drift, AI-era evidence, and identity claims.
  71. The Dating Agent Becomes the Receiver’s Opt-In Blog · Modified August 18, 2026, 11:44 PM PDT Critical analysis of arXiv:2608.18058v1 on send-receive receptivity, dating agents, stated-preference counterfactuals, and receiver choice.
  72. The Identical Post Becomes the Author Test Blog · Modified August 18, 2026, 8:30 PM PDT Critical analysis of arXiv:2608.13879v1 on identical-text exposure gaps, author history, and the limits of decentralized feed choice.
  73. Platform Monopoly Power Wiki · Modified August 18, 2026, 7:44 PM PDT Platform monopoly power: market-specific durable control over digital gateways, current U.S., EU, and UK proceedings, AI-stack risks, remedies, and evidence records.
  74. Programmed Inequality and the Labor Hidden Inside Computing Book Review · Modified August 18, 2026, 7:44 PM PDT A critical review of Mar Hicks's Programmed Inequality: gendered job classification, classification debt, technical capability, AI automation, worker power, governance, and safety.
  75. The Theory Loop Becomes the Cognitive Scientist Blog · Modified August 18, 2026, 7:43 PM PDT A source-disciplined analysis of AutoCog, closed-loop cognitive theory search, prospective testing, human-subject safeguards, and the governance record required for automated experiments.
  76. The Agent Endpoint Becomes the Unfinished Trial Blog · Modified August 18, 2026, 5:18 PM PDT Critical analysis of arXiv:2608.14940v1 on outcome finality, cross-unit separation, delayed effects, and open-effects records in agent evaluation.
  77. The Public-Service Agent Becomes the Queue Multiplier Blog · Modified August 18, 2026, 2:04 PM PDT Critical analysis of arXiv:2608.16603v1 on agent-assisted government requests, administrative burden, service capacity, friction, and procedural equality.
  78. The Safety Report Becomes the Neighborhood Border Blog · Modified August 18, 2026, 10:53 AM PDT Critical analysis of arXiv:2608.12581v1 on geolocated citizen safety reports, migrant-linked hate classification, spatial aggregation, platform engagement, and place-based moderation.
  79. The Grounding Gap Becomes the Backdoor Review Queue Book Review · Modified August 18, 2026, 7:36 AM PDT Critical analysis of arXiv:2608.12623v1 on Groundedness Drift, explanation-bearing language model classifiers, black-box backdoor auditing, and review routing.
  80. OpenID Connect Native SSO Wiki · Modified August 18, 2026, 6:51 AM PDT OpenID Connect Native SSO reference: same-vendor mobile session reuse, device-secret token exchange, draft status, validation, platform boundaries, privacy, and governance.
  81. The Pronunciation Correction Becomes the Voice Memory Blog · Modified August 18, 2026, 6:50 AM PDT Critical analysis of FlowEdit pronunciation memory: what its benchmark establishes, what a latent correction stores, and how consent, scope, deletion, and audit should work.
  82. When Old Technologies Were New and the AI Etiquette Panic Book Review · Modified August 18, 2026, 6:48 AM PDT A critical review of Carolyn Marvin's When Old Technologies Were New: social distance, expert authority, AI disclosure, provenance, consent, and etiquette becoming governance.
  83. The Privacy Notice Becomes the Release Artifact Blog · Modified August 18, 2026, 4:21 AM PDT Critical analysis of arXiv:2608.12511v1 on privacy documents as evolving software components, code traceability, cross-source consistency, AI systems, and release evidence.
  84. The AI Disclosure Becomes the Production Credit Blog · Modified August 18, 2026, 1:10 AM PDT Critical analysis of arXiv:2608.11539v1 on Steam reviews, generative-AI disclosures, player reception, creative labor, production signals, and evidentiary limits.
  85. The Synthetic Post Becomes the Containment Boundary Blog · Modified August 17, 2026, 10:06 PM PDT Critical analysis of arXiv:2608.10115v1 on Outer Limits, controlled Reddit experiments, interface fidelity, contained interactions, deception, and research governance.
  86. The Privacy Law Becomes the Household Alibi Blog · Modified August 17, 2026, 6:53 PM PDT Critical analysis of arXiv:2608.13582v1 on Jordanian smart homes, domestic workers, privacy law, household surveillance, consent, power, and the compliance paradox.
  87. The Patient Portal Reply Becomes the Clinical Voice Blog · Modified August 17, 2026, 5:58 PM PDT A governance analysis of AI-assisted patient portal communication: routing, chart retrieval, drafting, clinical review, disclosure, privacy, safety evidence, and accountable sending.
  88. Activation Steering Wiki · Modified August 17, 2026, 5:53 PM PDT A sourced reference on activation steering: runtime interventions in model activations, method boundaries, current evidence, reliability limits, safety risks, and governance controls.
  89. Cloud Empires and the Platform as Private Sovereign Book Review · Modified August 17, 2026, 5:52 PM PDT A critical review of Vili Lehdonvirta's Cloud Empires: platform sovereignty, cloud lock-in, algorithmic labor, AI infrastructure, contestability, appeal, and tested exit.
  90. The Tap Trace Becomes the Delegation Claim Blog · Modified August 17, 2026, 3:27 PM PDT Critical analysis of arXiv:2608.14132v1 on Act2Intention, mobile GUI action traces, user-intention prediction, synthetic personas, proactive agents, confirmation, and delegated authority.
  91. The Delusion Prompt Becomes the Safety Trajectory Blog · Modified August 17, 2026, 12:14 PM PDT Critical analysis of arXiv:2608.13017v1 on escalating delusion prompts, longitudinal model behavior, recognition timing, safeguarding, and trajectory-based AI safety evaluation.
  92. The General Assistant Becomes the Relational Actor Blog · Modified August 17, 2026, 9:00 AM PDT Critical analysis of arXiv:2608.10672v1 on general-purpose chatbot self-disclosure, relational prompting, felt closeness, and behavior-based governance.
  93. The Journal Entry Becomes the Follow-Through Score Blog · Modified August 17, 2026, 5:45 AM PDT Critical analysis of arXiv:2608.12582v1 on AI-supported journaling, passive sensing, intention classification, behavioral follow-through, and causal limits.
  94. The Eye of the Master and the Labor Hidden Inside AI Book Review · Modified August 17, 2026, 5:03 AM PDT A critical review of Matteo Pasquinelli's The Eye of the Master: labor as AI infrastructure, supervisory asymmetry, algorithmic management, worker power, and accountable automation.
  95. The Binary Question Becomes the Evaluation Probe Blog · Modified August 17, 2026, 5:01 AM PDT Critical analysis of arXiv:2606.27226 on BinEval: atomic binary evaluation questions, benchmark evidence, prompt-update leakage, judge security, and governance receipts.
  96. AI Bill of Materials Wiki · Modified August 17, 2026, 5:00 AM PDT AI bill of materials explained: scope, component identity, dependency graphs, CycloneDX and SPDX support, evidence integrity, governance risks, and lifecycle controls.