{ "title": "AI Thesis Atlas", "asOf": "2026-10-31", "scope": "A curated selection of 25 distinct AI perspectives: 10 venture-capital views, 8 finance, economics, labor and energy views, and 7 media, education, creativity, science and public-interest views. This is not an exhaustive archive. Sources include essays, reports, policies, research papers and an investment announcement; they are not all slide presentations.", "editorialNote": "Summaries report the cited source claims. Mechanisms, assumptions, counterarguments, falsification tests, practical implications, visual concepts, intern exercises, skills and proof-of-work requirements are original atlas analysis unless expressly labeled otherwise. They do not imply source endorsement of IndustryNext, ElSegundo.xyz, PointCast or UES. Forecasts, estimates, policy frameworks and investor opinions are labeled separately from empirical findings. Investors and other institutions have interests that may shape their perspectives.", "learningNote": "Exercises are proposed learning projects, not active internships, employer partnerships, job offers or placement promises. Use public or clearly licensed materials, obtain consent for any participants, minimize personal data, and document limitations. No applications or personal information are collected by this dataset.", "internResourceUrl": "https://elsegundo.xyz/property/#internships", "visualRules": [ "Retain source units, population, geography, time horizon and publication vintage beside every number.", "Label illustrative scenarios and synthetic diagrams as editorial illustrations; they are not measured outcomes, likelihoods or investor forecasts.", "Do not compare unlike units on one numeric axis, sum totals with subsets, or treat exposure as job loss.", "Keep source-reported values separate from editable hypothetical inputs.", "Provide equivalent text and table content for every visual; do not rely on color alone." ], "flatiron": { "status": "Unresolved identity; not counted among the 25 perspectives", "finding": "No clearly attributable public VC AI thesis from the requested Flatiron was verified. Do not replace it with Flat Capital or present Flatiron Health as a VC.", "primaryCandidates": [ { "title": "Scaling insights in cancer care with machine learning", "entity": "Flatiron Health", "date": "2022-12-02", "url": "https://resources.flatiron.com/scaling-insights-in-cancer-care-with-machine-learning", "note": "Public research presentation with transcript; healthcare company, not the verified intended VC." }, { "title": "Flatiron Partners", "entity": "UBS Private Wealth Management team", "url": "https://advisors.ubs.com/flatiron/", "note": "Name collision; no relevant AI thesis established." } ], "clarificationNeeded": "A Flatiron firm URL, author or deck title is needed before assigning a view." }, "accessNotes": [ { "target": "https://www.jpmorgan.com/insights/global-research/artificial-intelligence/ai-capex", "status": "excluded", "reason": "HTTP 403 during research; no record included and no bypass attempted." }, { "target": "https://www.nber.org/papers/w32487 and https://www.nber.org/papers/w31161", "status": "alternate primary sources used", "reason": "HTTP 403 during research; accessible author and institution primary sources were used." }, { "target": "https://dspace.mit.edu/entities/publication/0fcf7575-a092-4728-a0c5-077beda38427", "status": "alternate primary source used", "reason": "HTTP 429 during research; accessible MIT-hosted PDF was used." } ], "entries": [ { "id": "vc-value-capture", "title": "Creating value is not the same as capturing it", "institution": "Andreessen Horowitz (a16z)", "authors": "Matt Bornstein; Guido Appenzeller; Martin Casado", "date": "2023-01-19", "sourceTitle": "Who Owns the Generative AI Platform?", "url": "https://a16z.com/who-owns-the-generative-ai-platform/", "category": "Venture capital", "evidenceType": "Investor analytical opinion; early market observations", "summary": "Creating AI value and capturing its profits are different problems. Infrastructure initially benefits while apps and models must establish durable differentiation.", "mechanism": "Inference costs flow upstream; retention, workflow depth and distribution determine where margins remain.", "assumptions": "Comparable models remain available; customers can switch undifferentiated apps.", "counterargument": "A dominant model or a deeply embedded application could retain pricing power.", "falsifier": "Sustained high margins and retention in easily replicated apps would weaken the defensibility concern.", "implication": "Assess retention and contribution margin, not just revenue growth.", "internExercise": "Map the infrastructure, model and application dependencies of one public AI product. Draft a cost ledger with documented prices where available and explicitly unknown values elsewhere; identify a workflow that would make switching costly.", "skills": [ "Value-chain analysis", "Source verification", "Unit economics" ], "proofOfWork": "An annotated dependency map and cost ledger that distinguish observed prices, unknown inputs and original hypotheses.", "visual": { "type": "steps", "concept": "Clickable stack: infrastructure → models → apps. Toggle commoditization versus workflow lock-in; arrows describe mechanisms, not measured money flows.", "values": [] }, "caveat": "A January 2023 investor perspective, not a current measurement of margins or market power. Mechanism labels, counterarguments, falsifiers, practical implications and exercises are atlas analysis. The firm has financial interests in AI outcomes." }, { "id": "vc-labor-budget", "title": "Sell accepted work, then count the rework", "institution": "Andreessen Horowitz (a16z)", "authors": "Alex Rampell", "date": "2024-08-22", "sourceTitle": "Input Coffee, Output Code: How AI Will Turn Capital into Labor", "url": "https://a16z.com/ai-turns-capital-to-labor/", "category": "Venture capital", "evidenceType": "Prediction and business-model thesis", "summary": "Rampell argues that AI may let software vendors charge for work performed, expanding software markets beyond per-seat pricing.", "mechanism": "Agents act on stored records; usage or outcome pricing replaces dependence on human seat counts.", "assumptions": "Reliable execution; affordable inference and oversight; customers accept delegated work.", "counterargument": "Exceptions, liability and human review can erase the apparent savings.", "falsifier": "Outcome-priced products repeatedly fail to outperform human workflows after all oversight costs.", "implication": "Pilot a bounded task and measure accepted outcomes, rework and total cost.", "internExercise": "Choose a low-stakes task such as checking public opening hours. Define an acceptance checklist, compare manual and AI-assisted draft workflows, and log research time, review time and corrections without publishing unreviewed outputs.", "skills": [ "Workflow design", "Quality assurance", "Cost accounting" ], "proofOfWork": "A reproducible task brief, acceptance checklist and anonymized time-and-rework log; no claimed savings without measurements.", "visual": { "type": "scenario", "concept": "Seat versus outcome pricing switch, with editable hypothetical task volume, review burden and cost. Label all inputs illustrative; never label payroll as automatically addressable revenue.", "values": [] }, "caveat": "A business-model prediction, not proof that an entire labor budget becomes vendor revenue. Any calculator inputs are hypothetical and outputs are scenarios, not forecasts. The firm has financial interests in AI outcomes." }, { "id": "vc-infrastructure-payback", "title": "Infrastructure spending still needs a payback story", "institution": "Sequoia Capital", "authors": "David Cahn", "date": "2024-06-20", "sourceTitle": "AI’s $600B Question", "url": "https://sequoiacap.com/article/ais-600b-question", "category": "Venture capital", "evidenceType": "Dated investor scenario, not realized revenue or a current forecast", "summary": "The infrastructure buildout implies an enormous revenue requirement; technological usefulness does not guarantee attractive returns on the assets funding it.", "mechanism": "Convert chip revenue into total datacenter costs, then into customer revenue needed at an assumed margin.", "assumptions": "GPU costs represent half of total ownership costs; downstream gross margin is 50%.", "counterargument": "Assets produce revenue over multiple years; strategic spending and productivity gains need not appear as new AI-vendor revenue.", "falsifier": "Durable monetization and returns support the buildout without the assumed revenue gap.", "implication": "Stress-test utilization, price compression and depreciation separately.", "internExercise": "Rebuild Cahn’s dated heuristic in a spreadsheet, clearly mark its assumptions, and add a separate sensitivity tab for utilization and useful asset life without presenting the extension as Cahn’s model.", "skills": [ "Scenario modeling", "Spreadsheet audit", "Capital-cost analysis" ], "proofOfWork": "A source-linked workbook with separate reproduced and illustrative tabs, unit labels and an assumptions note.", "visual": { "type": "scenario", "concept": "Recreate the dated assumption waterfall with editable multipliers. Clearly label it Cahn’s June 2024 heuristic, not audited breakeven accounting. Keep unlike units separate.", "values": [ { "label": "Datacenter cost multiplier", "value": 2, "unit": "x", "kind": "author assumption", "status": "author assumption", "year": 2024, "sourceUrl": "https://sequoiacap.com/article/ais-600b-question", "basis": "source-reported", "role": "assumption" }, { "label": "Downstream gross margin", "value": 50, "unit": "percent", "kind": "author assumption", "status": "author assumption", "year": 2024, "sourceUrl": "https://sequoiacap.com/article/ais-600b-question", "basis": "source-reported", "role": "assumption" }, { "label": "Implied annual revenue requirement", "value": 600, "unit": "USD billion", "kind": "June 2024 scenario output", "status": "June 2024 scenario output", "year": 2024, "sourceUrl": "https://sequoiacap.com/article/ais-600b-question", "basis": "source-reported", "role": "output" } ] }, "caveat": "Cahn’s June 2024 investor heuristic is not audited breakeven accounting, realized revenue or a current forecast. The 2× cost multiplier and 50% margin are author assumptions; the $600 billion figure is the dated scenario output. No investment recommendation." }, { "id": "vc-consumer-personalization", "title": "Personalization must become repeat utility", "institution": "Andreessen Horowitz (a16z)", "authors": "Connie Chan; Justine Moore", "date": "2023-02-07", "sourceTitle": "Generative AI: The Next Consumer Platform", "url": "https://a16z.com/generative-ai-the-next-consumer-platform/", "category": "Venture capital", "evidenceType": "Prediction and opportunity map", "summary": "Personalized generation can create consumer categories and improve existing experiences across search, education and creation.", "mechanism": "Natural-language interaction and adaptive content reduce the effort required to get a tailored result.", "assumptions": "Novelty converts into repeat utility; personalization is trustworthy and economical.", "counterargument": "General assistants or incumbent distribution may absorb standalone products; low switching costs weaken loyalty.", "falsifier": "Usage falls after novelty wears off and personalization does not improve retained utility.", "implication": "Test repeat use for a concrete consumer problem before treating downloads as product-market fit.", "internExercise": "Prototype generic and user-tailored versions of a local reading guide using public sources. Conduct a consented usability exercise about usefulness and repeat intent, collecting only information needed for the test.", "skills": [ "Interaction design", "Usability research", "Data minimization" ], "proofOfWork": "Two clickable prototypes, an opt-in testing script and a report that separates observed behavior from stated intent.", "visual": { "type": "scenario", "concept": "Choose a consumer task and toggle generic versus adaptive experience; reveal accuracy, privacy and retention trade-offs without invented scores.", "values": [] }, "caveat": "A February 2023 opportunity map, not current consumer retention data. Prototype examples and trade-offs are editorial illustrations. The firm has financial interests in AI outcomes." }, { "id": "vc-workflow-replacement", "title": "Systems that act could challenge systems that record", "institution": "Bessemer Venture Partners", "authors": "Mike Droesch; Darsh Patel", "date": "2025-05-29", "sourceTitle": "Roadmap: AI systems of action", "url": "https://www.bvp.com/atlas/roadmap-ai-systems-of-action", "category": "Venture capital", "evidenceType": "Investor prediction", "summary": "AI may weaken incumbent systems-of-record moats by lowering migration friction while making software perform workflows rather than merely store them.", "mechanism": "Models structure messy data, generate integrations and automate actions.", "assumptions": "Migration becomes materially safer and cheaper; automation benefits exceed switching risk.", "counterargument": "Incumbents own trusted data, procurement relationships and distribution, and can add the same models.", "falsifier": "Migration remains expensive while incumbent renewal and workflow control remain intact.", "implication": "Build a narrow action wedge with a credible migration and audit plan.", "internExercise": "Use synthetic business records to design one reversible action workflow. Map field transformations, approval gates and a rollback procedure; test malformed records before proposing a live integration.", "skills": [ "Data modeling", "Workflow prototyping", "Audit-trail design" ], "proofOfWork": "A synthetic-data prototype with a migration map, failure cases, approval diagram and demonstrated rollback.", "visual": { "type": "steps", "concept": "Interactive migration journey: data cleaning → integration → action → record ownership. Toggle incumbent response and compliance burden; show qualitative consequences.", "values": [] }, "caveat": "An investor prediction about competitive structure, not evidence that migration is safe or cheap in every organization. Journey steps and failure paths are editorial interpretations." }, { "id": "vc-defensibility", "title": "Early momentum is not a permanent moat", "institution": "NFX", "authors": "Pete Flint", "date": "2025-07", "datePrecision": "month", "sourceTitle": "How AI Companies Will Build Real Defensibility", "url": "https://www.nfx.com/post/ai-defensibility", "category": "Venture capital", "evidenceType": "Strategic opinion", "summary": "Early distribution and speed are not permanent moats. AI companies should use initial momentum to develop embedding, switching costs and genuine network effects.", "mechanism": "More shared context or collaborators can improve utility and deepen workflow dependence.", "assumptions": "Additional participation creates incremental value; context is difficult to replicate or port.", "counterargument": "Portable memory, interoperable agents and model convergence can reduce switching costs.", "falsifier": "Larger networks or accumulated context fail to improve retention or customer utility.", "implication": "Distinguish a growth loop from a network effect and test the incremental-user benefit.", "internExercise": "Audit a collaborative product using public information. Draw separate acquisition and utility loops, identify which claim would require another participant to improve the experience, and draft a consented test of that claim.", "skills": [ "Product analysis", "Experiment design", "Causal reasoning" ], "proofOfWork": "Two labeled loop diagrams and a test plan that states what observation would refute the proposed network effect.", "visual": { "type": "matrix", "concept": "Layered moat builder with separate cards for distribution, brand, embedding and network effects. No numeric strength rankings; user-selected assumptions expose failure paths.", "values": [] }, "caveat": "Strategic opinion; publication date is established only to July 2025. The visual must not assign unsupported moat-strength scores. The firm has financial interests in AI outcomes." }, { "id": "vc-enterprise-buying", "title": "Enterprise spending is growing—but spending is not ROI", "institution": "Menlo Ventures", "authors": "Tim Tully; Joff Redfern; Deedy Das; Derek Xiao", "date": "2025-12", "datePrecision": "month; exact publication day not established from opened primary page", "sourceTitle": "2025: The State of Generative AI in the Enterprise", "url": "https://menlovc.com/perspective/2025-the-state-of-generative-ai-in-the-enterprise/", "category": "Venture capital", "evidenceType": "Investor survey and modeled market estimates", "summary": "Menlo finds enterprise AI purchasing expanding beyond experiments, with applications receiving substantial spending and buyers increasingly choosing external products.", "mechanism": "Ready-made products reach production faster than internal builds.", "assumptions": "Survey and modeled private-company revenue represent the target market.", "counterargument": "AI-active U.S. respondents and investor portfolio interests can skew the picture; spending is not proof of ROI.", "falsifier": "Renewals, realized savings or independently measured adoption materially lag the spending narrative.", "implication": "Compare buying with building using deployment time and total ownership cost.", "internExercise": "Create a buy-versus-build decision memo for a fictional team. Define the task, data access, deployment effort and review costs; distinguish sourced vendor facts from assumptions and identify what a trial must measure.", "skills": [ "Procurement research", "Total-cost modeling", "Survey literacy" ], "proofOfWork": "A source-linked decision memo and assumptions workbook with a trial checklist; no unsupported ROI claim.", "visual": { "type": "bar", "concept": "Sourced spending bars with methodology drawer. Keep excluded chips, inference/serving and embedded features visible. Total is composed of applications and infrastructure; do not sum the total with its components. Present survey count separately from money.", "values": [ { "label": "2025 enterprise GenAI spending", "value": 37, "unit": "USD billion", "kind": "Menlo market estimate", "status": "Menlo market estimate", "year": 2025, "sourceUrl": "https://menlovc.com/perspective/2025-the-state-of-generative-ai-in-the-enterprise/", "basis": "source-reported", "group": "Spending estimate", "role": "total" }, { "label": "2025 applications", "value": 19, "unit": "USD billion", "kind": "Menlo rounded estimate", "status": "Menlo rounded estimate", "year": 2025, "sourceUrl": "https://menlovc.com/perspective/2025-the-state-of-generative-ai-in-the-enterprise/", "basis": "source-reported", "group": "Spending estimate", "role": "component" }, { "label": "2025 infrastructure", "value": 18, "unit": "USD billion", "kind": "Menlo rounded estimate", "status": "Menlo rounded estimate", "year": 2025, "sourceUrl": "https://menlovc.com/perspective/2025-the-state-of-generative-ai-in-the-enterprise/", "basis": "source-reported", "group": "Spending estimate", "role": "component" }, { "label": "Survey respondents", "value": 495, "unit": "U.S. enterprise AI decision-makers", "kind": "survey sample; November 7–25, 2025", "status": "survey sample; November 7–25, 2025", "year": 2025, "sourceUrl": "https://menlovc.com/perspective/2025-the-state-of-generative-ai-in-the-enterprise/", "basis": "source-reported", "group": "Survey sample", "role": "metadata" } ] }, "caveat": "Menlo’s market definition excludes chips, inference/serving and embedded features; it is not all AI spending. Estimates and rounded components are not audited totals. Survey sample: 495 U.S. enterprise AI decision-makers, November 7–25, 2025. Exact publication day was not established. Spending does not prove ROI; portfolio interests may affect framing." }, { "id": "vc-abundance-distribution", "title": "Abundance also depends on who receives the gains", "institution": "Khosla Ventures", "authors": "Vinod Khosla", "date": "2024-11-11", "sourceTitle": "A Roadmap to AI Utopia", "url": "https://www.khoslaventures.com/posts/a-roadmap-to-ai-utopia", "category": "Venture capital", "evidenceType": "Normative opinion and long-horizon prediction", "summary": "Abundant machine intelligence could reduce scarcity, but broad prosperity depends on policy that distributes gains and supports displaced workers.", "mechanism": "Scalable cognitive work lowers service costs; redistribution and transition support spread benefits.", "assumptions": "Capability translates into productivity; institutions can manage transition and preserve human agency.", "counterargument": "Concentrated ownership, scarce physical inputs and political resistance may prevent broad abundance.", "falsifier": "Sustained capability growth fails to lower essential-service costs or widen access.", "implication": "Evaluate distribution and affordability alongside aggregate output.", "internExercise": "Choose a public-facing service and map who pays, who works and who benefits. Write contrasting scenarios for lower production costs with broad versus concentrated access, identifying evidence needed to distinguish them.", "skills": [ "Stakeholder mapping", "Policy analysis", "Scenario writing" ], "proofOfWork": "A stakeholder map and two clearly hypothetical service-access scenarios with measurable questions and no assigned probabilities.", "visual": { "type": "matrix", "concept": "Two-axis scenario map: productivity gains versus distribution choices. Show qualitative futures; explicitly call scenarios editorial interpretations, not probabilities.", "values": [] }, "caveat": "A normative, long-horizon vision rather than an empirical forecast. Scenario quadrants are editorial interpretations and have no assigned probabilities. No source endorsement of the atlas or its exercises is implied." }, { "id": "vc-ai-rollups", "title": "Owning a service business changes the AI bet", "institution": "General Catalyst", "authors": "Marc Bhargava; Kate Bender", "date": "2025-08-28", "sourceTitle": "The Future of Services", "url": "https://www.generalcatalyst.com/stories/the-future-of-services", "category": "Venture capital", "evidenceType": "Investment thesis; portfolio examples and targets, not independent causal evidence", "summary": "Owning service businesses can accelerate AI adoption and capture the value of improved outcomes rather than only selling software licenses.", "mechanism": "Acquire customers and operational workflows, integrate domain AI, expand service capacity, then repeat.", "assumptions": "Integration works; customers stay; technology savings exceed acquisition and transformation costs.", "counterargument": "Operational complexity, debt and inconsistent data can make a roll-up worse than a focused software vendor.", "falsifier": "Acquired cohorts fail to improve organic growth, service quality and cash margins after integration costs.", "implication": "Separate acquired growth from organic improvement and track cohort-level quality.", "internExercise": "Build a fictional service-business operating dashboard that separately labels purchased revenue, organic customer growth, review cost and service quality. Use invented sample data only and trace every scenario output to its inputs.", "skills": [ "Operational analysis", "Cohort tracking", "Data visualization" ], "proofOfWork": "A synthetic dashboard and data dictionary that prevent acquired revenue from being mistaken for organic improvement.", "visual": { "type": "steps", "concept": "Acquisition-to-improvement loop with a toggle separating organic growth, purchased revenue and operational savings. Illustrative scenarios only, not valuation advice.", "values": [] }, "caveat": "An investor thesis using portfolio examples and targets, not independent causal evidence. Synthetic examples must remain labeled illustrative. No valuation or investment recommendation." }, { "id": "vc-security-agents", "title": "Security agents need approval gates and a way back", "institution": "Greylock", "authors": "Saam Motamedi", "date": "2025-07-14", "sourceTitle": "Introducing Cogent: AI Agents for Vulnerability Management", "url": "https://greylock.com/blog/introducing-cogent-ai-agents-for-vulnerability-management/", "category": "Venture capital", "evidenceType": "Investment announcement and portfolio-product thesis; promotional claims", "summary": "Greylock presents a product thesis: agents could identify, prioritize and remediate vulnerabilities, moving security teams beyond alert queues. This investment announcement is not independent evidence of performance.", "mechanism": "Agents connect asset context, vulnerability assessment and execution rather than only surfacing alerts.", "assumptions": "Remediation is reliable; permissions are controlled; telemetry is accurate.", "counterargument": "An autonomous security tool can introduce outages or become a privileged attack surface.", "falsifier": "Audited deployments show no net improvement after false positives, rollbacks and incident costs.", "implication": "Start with bounded permissions, human approval and reversible changes.", "internExercise": "Create a tabletop exercise with synthetic vulnerability tickets. Design triage, reviewer approval and rollback states, then walk through a false positive and a failed remediation without scanning or changing any real system.", "skills": [ "Threat modeling", "Permission design", "Incident documentation" ], "proofOfWork": "A synthetic ticket set, state diagram, approval checklist and recorded tabletop outcomes including failure recovery.", "visual": { "type": "steps", "concept": "Clickable vulnerability lifecycle with human approval gates and a rollback branch. Label it a proposed architecture; omit unverified performance benchmarks.", "values": [] }, "caveat": "A promotional investment announcement, not independently verified product-performance evidence. The proposed architecture and tabletop are editorial learning tools, not operational security instructions." }, { "id": "finance-goldman-measurement", "title": "The AI investment boom depends on what gets counted", "institution": "Goldman Sachs Research", "authors": "Institutional article reporting Joseph Briggs's research", "date": "2026-08-07", "sourceTitle": "Global AI Investment Is Forecast to Exceed $1 Trillion in 2026", "url": "https://www.goldmansachs.com/insights/articles/global-investment-is-forecast-to-exceed-1-trillion-in-2026", "category": "Finance & economy", "field": "Traditional finance", "evidenceType": "Institutional forecast, not realized spending", "summary": "Goldman estimates roughly $1 trillion of global AI investment in 2026, including $581 billion in the United States. Hyperscaler capex alone is neither a complete global total nor a clean measure of domestic AI spending.", "mechanism": "Add non-hyperscaler and foreign investment; adjust non-AI baseline spending, geography and leases.", "assumptions": "Incremental capex above 2022 is largely AI-related; project geography approximates actual investment location.", "counterargument": "The baseline can misclassify ordinary cloud expansion; opaque lease disclosures permit double counting.", "falsifier": "Reconcile final corporate capex and government investment data with consistent AI definitions; large unexplained residuals undermine the estimate.", "implication": "Define the spending denominator before presenting an AI boom chart.", "internExercise": "Build an auditable capex-classification sheet from public filings; flag leases, geography and unknowns.", "skills": [ "Financial source reading", "Data classification", "Chart annotation" ], "proofOfWork": "A source-linked spreadsheet with one row per filing, a reproducible inclusion rule, and an annotated chart that keeps US spending within the global total.", "visual": { "type": "bar", "concept": "Toggle global versus US estimates with methodology cards; never add overlapping totals.", "values": [ { "label": "Global AI investment", "value": 1000, "unit": "USD billions", "year": 2026, "status": "forecast, approximately", "sourceUrl": "https://www.goldmansachs.com/insights/articles/global-investment-is-forecast-to-exceed-1-trillion-in-2026", "basis": "source-reported" }, { "label": "US AI investment", "value": 581, "unit": "USD billions", "year": 2026, "status": "forecast, subset of global", "sourceUrl": "https://www.goldmansachs.com/insights/articles/global-investment-is-forecast-to-exceed-1-trillion-in-2026", "basis": "source-reported" } ] }, "caveat": "These are August 2026 institutional forecasts, not realized spending. The US estimate is a subset of the global estimate, not an additional amount. The source's AI definition and baseline are part of the estimate. Educational analysis, not an investment recommendation." }, { "id": "finance-blackrock-scarcity", "title": "AI abundance may arrive through a period of scarcity", "institution": "BlackRock Investment Institute", "authors": "BlackRock Investment Institute (institutional authorship)", "date": "2026-09-15", "sourceTitle": "Q4 Global Investment Outlook: Scarcity vs. abundance", "url": "https://www.blackrock.com/us/individual/insights/blackrock-investment-institute/outlook", "category": "Finance & economy", "field": "Traditional finance", "evidenceType": "Institutional outlook, not a causal estimate", "summary": "BlackRock argues that AI investment supports activity while absorbing capital, power, materials and balance-sheet capacity. Its outlook highlights financing and capacity constraints alongside the prospect of future productivity gains.", "mechanism": "Front-loaded construction and equipment demand meet limited capacity and competing sovereign borrowing.", "assumptions": "Buildout continues; constrained inputs cannot expand quickly; cash generation lags some financing needs.", "counterargument": "Efficiency improvements or weaker demand could release capacity, while cheaper financing could ease the bottleneck.", "falsifier": "Track project delays, refinancing terms and utilization together; abundant capacity with sustained demand weakens the scarcity thesis.", "implication": "Evaluate deployment dependencies and financing resilience, not only model capability.", "internExercise": "Create a risk register for a hypothetical data-center project with evidence links and mitigations.", "skills": [ "Dependency mapping", "Evidence evaluation", "Risk-register design" ], "proofOfWork": "A hypothetical project dependency map and risk register identifying a source, an unresolved question, and a potential mitigation for each constraint.", "visual": { "type": "steps", "concept": "Clickable dependency map: capital, power, materials, labor, revenue. Qualitative links only; no fabricated weights.", "values": [], "unit": "qualitative", "year": 2026, "nodes": [ "Capital", "Power", "Materials", "Labor", "Revenue" ] }, "caveat": "The public US outlook page mixes a September header with October subsections. This record uses the dated header thesis only; investment recommendations are omitted. The dependency map is an editorial explanation, not measured relationships or a prediction of project outcomes.", "verificationNote": "Public US page opened successfully. It mixes September header and October subsections; use dated header thesis only. Investment recommendations intentionally omitted." }, { "id": "economics-imf-exposure", "title": "Exposure is not the same thing as job loss", "institution": "International Monetary Fund", "authors": "Kristalina Georgieva", "date": "2024-01-14", "sourceTitle": "AI Will Transform the Global Economy. Let’s Make Sure It Benefits Humanity.", "url": "https://www.imf.org/en/blogs/articles/2024/01/14/ai-will-transform-the-global-economy-lets-make-sure-it-benefits-humanity", "category": "Finance & economy", "field": "Economics and labor", "evidenceType": "Exposure estimates and conditional policy outlook", "summary": "The IMF places almost 40% of global employment in AI-exposed work. Some exposed tasks may be complemented and others substituted; countries' ability to capture gains differs.", "mechanism": "Task changes interact with infrastructure, skills, labor mobility and ownership of capital.", "assumptions": "Occupational exposure captures meaningful future task change; institutions shape adaptation.", "counterargument": "Exposure scores cannot determine adoption speed, net employment or wages. New demand and jobs may offset substitution.", "falsifier": "Compare actual task adoption and worker transitions with exposure classifications; persistent mismatch challenges predictive usefulness.", "implication": "Design training around tasks and worker transitions rather than treating exposure as a redundancy forecast.", "internExercise": "Map tasks in one occupation to complement, substitute or uncertain categories, then interview practitioners with consent.", "skills": [ "Task analysis", "Consent-based interviewing", "Statistical interpretation" ], "proofOfWork": "A task map with evidence and uncertainty labels, interview-consent notes, and a short explanation of why exposure cannot be converted into a job-loss forecast.", "visual": { "type": "bar", "concept": "Exposure bars by economy group with a permanent 'not job-loss probabilities' label.", "values": [ { "label": "Advanced economies", "value": 60, "unit": "percent of employment exposed", "year": 2024, "status": "estimate, about", "sourceUrl": "https://www.imf.org/en/blogs/articles/2024/01/14/ai-will-transform-the-global-economy-lets-make-sure-it-benefits-humanity", "basis": "source-reported" }, { "label": "Emerging markets", "value": 40, "unit": "percent of employment exposed", "year": 2024, "status": "estimate", "sourceUrl": "https://www.imf.org/en/blogs/articles/2024/01/14/ai-will-transform-the-global-economy-lets-make-sure-it-benefits-humanity", "basis": "source-reported" }, { "label": "Low-income countries", "value": 26, "unit": "percent of employment exposed", "year": 2024, "status": "estimate", "sourceUrl": "https://www.imf.org/en/blogs/articles/2024/01/14/ai-will-transform-the-global-economy-lets-make-sure-it-benefits-humanity", "basis": "source-reported" } ] }, "caveat": "Exposure is not a probability of job loss, a redundancy forecast, or a measure of completed adoption. These January 2024 estimates describe task exposure; complementarity, substitution, institutions and new demand affect eventual outcomes." }, { "id": "finance-bis-systemic", "title": "Shared AI can create shared financial failure modes", "institution": "Bank for International Settlements", "authors": "BIS (institutional chapter)", "date": "2024-06-30", "sourceTitle": "Artificial intelligence and the economy: implications for central banks", "url": "https://www.bis.org/publications/aer-2024/artificial-intelligence-economy-implications-central-banks", "category": "Finance & economy", "field": "Traditional finance and public institutions", "evidenceType": "Risk mechanisms and institutional analysis, not crisis prediction", "summary": "BIS identifies a tension: AI can improve analysis and supervision, but dependence on common providers and algorithms can synchronize financial behavior and operational failures.", "mechanism": "Shared models, data and infrastructure connect institutions that may appear independent.", "assumptions": "Adoption becomes concentrated and decisions react similarly under stress.", "counterargument": "Diverse models, human controls and different objectives may prevent synchronized responses.", "falsifier": "Stress tests showing genuinely independent behavior and effective provider failover would weaken the concentration-risk mechanism.", "implication": "Treat model dependencies and exit options as operational-risk questions.", "internExercise": "Use synthetic institutions to document failure propagation and design a model-vendor exit checklist.", "skills": [ "Systems mapping", "Scenario design", "Operational-risk documentation" ], "proofOfWork": "A labeled synthetic dependency diagram with common-provider and diversified cases, plus a failover and exit checklist with explicit assumptions.", "visual": { "type": "scenario", "concept": "Dependency network with a toggle between diversified and common provider; explicitly hypothetical topology, not real exposure data.", "values": [], "unit": "qualitative", "year": 2024, "states": [ "Common provider", "Diversified providers" ] }, "caveat": "The network is hypothetical and contains no actual bank exposures or estimated crisis probabilities. Publication date comes from the parent annual-report listing. The source describes risk mechanisms rather than forecasting a crisis.", "verificationNote": "Chapter and parent annual report opened; publication date comes from parent annual-report listing." }, { "id": "economics-acemoglu-modest", "title": "Large task gains need not imply a huge macro boom", "institution": "Massachusetts Institute of Technology", "authors": "Daron Acemoglu", "date": "2024-05-12", "sourceTitle": "The Simple Macroeconomics of AI", "url": "https://economics.mit.edu/sites/default/files/2024-05/The%20Simple%20Macroeconomics%20of%20AI.pdf", "category": "Finance & economy", "field": "Academic economics", "evidenceType": "Conditional model-based projection; historical 2024 paper", "summary": "Acemoglu's task-based calculation puts US total-factor-productivity gains at no more than 0.66% over ten years under its baseline assumptions, and below 0.53% after harder-to-learn tasks are considered.", "mechanism": "Aggregate gains depend on economically relevant task shares and net task-level cost savings.", "assumptions": "Near-term benefits mainly come from existing-task automation or complementarity, not revolutionary scientific discoveries.", "counterargument": "New tasks, major scientific advances or cheaper capable systems could make the modeled channels too narrow.", "falsifier": "Broad, attributable productivity growth or large new-task contributions beyond the modeled range challenge the calibration.", "implication": "Separate macro forecasts from demonstrations of speed on selected tasks.", "internExercise": "Rebuild a toy task-share × cost-saving calculation and disclose every invented input as a scenario, not evidence.", "skills": [ "Spreadsheet modeling", "Assumption disclosure", "Economic measurement" ], "proofOfWork": "An editable toy model with a formula glossary, separately marked invented inputs, and a caption distinguishing cumulative TFP from annual growth and labor productivity.", "visual": { "type": "bar", "concept": "Two annotated conditional bounds; no annual growth-rate interpretation and no direct equivalence to labor productivity.", "values": [ { "label": "Baseline upper bound", "value": 0.66, "unit": "percent cumulative US TFP gain over ten years", "year": 2024, "status": "conditional projection; no more than this value", "sourceUrl": "https://economics.mit.edu/sites/default/files/2024-05/The%20Simple%20Macroeconomics%20of%20AI.pdf", "basis": "source-reported" }, { "label": "Hard-task-adjusted bound", "value": 0.53, "unit": "percent cumulative US TFP gain over ten years", "year": 2024, "status": "strictly below this value", "sourceUrl": "https://economics.mit.edu/sites/default/files/2024-05/The%20Simple%20Macroeconomics%20of%20AI.pdf", "basis": "source-reported" } ] }, "caveat": "These are cumulative, conditional US total-factor-productivity bounds over ten years from a historical 2024 model. They are not annual growth rates, measured realized gains, labor-productivity estimates, or a current comprehensive forecast of every possible AI channel." }, { "id": "labor-brynjolfsson-learning", "title": "AI assistance can spread expertise unevenly", "institution": "Stanford/MIT researchers; author-deposited research", "authors": "Erik Brynjolfsson, Danielle Li, Lindsey Raymond", "date": "2024-11-06", "sourceTitle": "Generative AI at Work (arXiv v2)", "url": "https://arxiv.org/abs/2304.11771v2", "category": "Finance & economy", "field": "Academic economics and labor", "evidenceType": "Empirical study of a staggered workplace rollout, not universal forecast", "summary": "In the revised study of 5,172 customer-support agents, AI assistance raised issues resolved per hour by 15% on average. Less experienced workers benefited more; experienced workers saw small speed gains and small quality declines.", "mechanism": "Assistance makes useful patterns available during work and may accelerate learning.", "assumptions": "The rollout permits credible comparisons; measured outcomes reflect meaningful productivity.", "counterargument": "One support setting does not establish effects in other occupations, long-run skill formation or employment.", "falsifier": "Replications with quality-adjusted outcomes and follow-up after tool removal can test generalization and durable learning.", "implication": "Measure quality and skill-specific effects, not just average speed.", "internExercise": "Run a consent-based practice-task evaluation measuring completion time, accuracy and retained skill.", "skills": [ "Evaluation design", "Quality measurement", "Descriptive analysis" ], "proofOfWork": "A small practice-task protocol, consent record, anonymized results table, and limitations note distinguishing a workshop demonstration from a replication.", "visual": { "type": "bar", "concept": "A single study-estimate bar plus qualitative experience-group cards; no fabricated subgroup percentages.", "values": [ { "label": "Average issues resolved per hour", "value": 15, "unit": "percent increase", "year": 2024, "status": "study estimate in revised paper", "sourceUrl": "https://arxiv.org/abs/2304.11771v2", "basis": "source-reported" } ], "qualitativeGroups": [ { "label": "Less experienced workers", "finding": "Benefited more" }, { "label": "Experienced workers", "finding": "Small speed gains and small quality declines" } ] }, "caveat": "The November 2024 arXiv v2 reports 15% across 5,172 customer-support agents. Do not mix these with the 2023 version's 14% and 5,179 figures. Findings come from one support setting and do not establish universal productivity, employment, or long-run learning effects.", "verificationNote": "Version-specific URL prevents mixing 2023 figures (14%, 5,179) with 2024 revision (15%, 5,172)." }, { "id": "energy-iea-bottlenecks", "title": "The AI buildout is also an electricity-system buildout", "institution": "International Energy Agency", "authors": "IEA (institutional report)", "date": "2025-04-10", "sourceTitle": "Energy and AI — Executive summary", "url": "https://www.iea.org/reports/energy-and-ai/executive-summary", "category": "Finance & economy", "field": "Energy and industrial infrastructure", "evidenceType": "2025-vintage modeled outlook; not latest-year actuals", "summary": "IEA's 2025 outlook projects global data-center electricity use rising from an estimated 415 TWh in 2024 to around 945 TWh in 2030. Local grid bottlenecks matter even when the global share remains modest.", "mechanism": "Compute demand drives electricity load; clustered development strains connections and supply chains.", "assumptions": "Demand, efficiency and deployment follow the report's modeled path; planned supply can be delivered.", "counterargument": "Efficiency, utilization changes or slower adoption could reduce demand; rebound could increase it.", "falsifier": "Compare metered data-center use, delivered capacity and connection delays against the modeled path.", "implication": "Assess local power availability before assuming compute capacity can scale.", "internExercise": "Build a hypothetical facility electricity budget using documented power and utilization assumptions.", "skills": [ "Unit conversion", "Energy-budget modeling", "Scenario documentation" ], "proofOfWork": "An auditable facility-energy worksheet with units, utilization assumptions and a sensitivity table; compare it with the report's context without treating the local scenario as a forecast.", "visual": { "type": "bar", "concept": "Estimated 2024 versus projected 2030 bars; clearly label all data centers, not AI-only demand.", "values": [ { "label": "2024 estimate", "value": 415, "unit": "TWh per year, global data-center electricity", "year": 2024, "status": "historical estimate", "sourceUrl": "https://www.iea.org/reports/energy-and-ai/executive-summary", "basis": "source-reported" }, { "label": "2030 projection", "value": 945, "unit": "TWh per year, global data-center electricity", "year": 2030, "status": "2025-vintage projection, around", "sourceUrl": "https://www.iea.org/reports/energy-and-ai/executive-summary", "basis": "source-reported" } ] }, "caveat": "Figures cover electricity for all data centers, not AI alone. The 2024 number is an estimate and the 2030 number is a projection from the April 2025 report; neither should be presented as the latest IEA forecast or latest-year actuals.", "verificationNote": "Report landing page confirms April 10, 2025. Retain vintage label rather than implying this is the newest IEA forecast." }, { "id": "labor-ilo-aggregation", "title": "Adoption and organizational change determine whether gains scale", "institution": "International Labour Organization", "authors": "Cheuk Yu Cheryl Chan; Khatia Shedania", "date": "2026-05-06", "sourceTitle": "The Aggregation Paradox of AI: Why do micro-economic productivity gains from AI disappear at scale", "url": "https://www.ilo.org/publications/aggregation-paradox-ai-why-do-micro-economic-productivity-gains-ai", "category": "Finance & economy", "field": "Labor institutions", "evidenceType": "Research synthesis and conditional institutional outlook", "summary": "The ILO brief argues that impressive task results have not yet clearly appeared as AI-driven aggregate productivity growth. Diffusion, workplace redesign, skills and institutions mediate the translation.", "mechanism": "Bottlenecks elsewhere in a workflow, complementary investment and uneven adoption absorb isolated efficiency gains.", "assumptions": "Current measurement captures enough activity to identify broad gains; organizational adaptation takes time.", "counterargument": "Official statistics may lag or miss new quality and output; aggregate acceleration could emerge abruptly.", "falsifier": "Sustained sector-wide gains linked to AI adoption, beyond adopters' task metrics, would qualify the diagnosis.", "implication": "Track end-to-end throughput and job quality alongside tool usage.", "internExercise": "Map one workflow and propose before/after measures for rework, throughput, quality and worker experience.", "skills": [ "Workflow mapping", "Measurement design", "Organizational analysis" ], "proofOfWork": "A workflow map and measurement plan with operational definitions, possible confounders, and a note explaining why a task-level gain cannot simply be scaled to the economy.", "visual": { "type": "steps", "concept": "Task → team → firm → economy funnel with clickable bottlenecks, no numeric conversion rates.", "values": [], "unit": "qualitative", "year": 2026, "nodes": [ "Task", "Team", "Firm", "Economy" ] }, "caveat": "This is a May 2026 research synthesis and conditional diagnosis, not proof that AI has no aggregate effect. The funnel is explanatory; it contains no measured conversion rates. Measurement lags and future adaptation remain open questions." }, { "id": "public-01", "title": "Convenience is becoming the news gateway", "institution": "Reuters Institute for the Study of Journalism, University of Oxford", "authors": "Felix M. Simon; Rasmus Kleis Nielsen; Richard Fletcher", "date": "2026-10-29", "sourceTitle": "Generative AI and News Report 2026: From standalone tools to information infrastructure", "url": "https://reutersinstitute.politics.ox.ac.uk/generative-ai-and-news-report-2026-standalone-tools-information-infrastructure", "category": "Media & learning", "evidenceType": "Nine-country online survey; six-country trend subset", "summary": "AI is becoming an information intermediary, but news remains a minority use. Across six comparable countries, weekly information-seeking grew from 11% in 2024 to 38% in 2026; news use grew from 3% to 9%.", "mechanism": "Conversational interfaces reduce effort and permit follow-up questions.", "assumptions": "Self-reports represent behavior; country composition and questionnaire framing matter.", "counterargument": "Convenience need not displace publishers. Survey sentiment alone does not establish substitution.", "falsifier": "Test actual referral, comprehension and repeat-use behavior rather than infer substitution from survey sentiment.", "implication": "PointCast should pair concise answers with visible original sources and follow-up paths.", "internExercise": "Prototype two sourced news-answer flows and run a consented usability exercise measuring source opening and factual recall; report sample limitations.", "skills": [ "Source evaluation", "Interaction design", "Usability research" ], "proofOfWork": "Two working answer-flow prototypes, a source ledger, and an anonymized observation report that separates source-opening behavior from factual recall and states sample limits.", "visual": { "type": "bar", "concept": "Toggle information-seeking and news across 2024, 2025 and 2026. Show the six-country weighted survey trend, not global adoption; keep the denominator and fieldwork caveat visible.", "values": [ { "label": "Information-seeking · 2024", "series": "Weekly information-seeking", "value": 11, "unit": "percent of survey respondents", "year": 2024, "status": "Six-country weighted survey trend", "sourceUrl": "https://reutersinstitute.politics.ox.ac.uk/generative-ai-and-news-report-2026-standalone-tools-information-infrastructure", "basis": "source-reported" }, { "label": "Information-seeking · 2025", "series": "Weekly information-seeking", "value": 24, "unit": "percent of survey respondents", "year": 2025, "status": "Six-country weighted survey trend", "sourceUrl": "https://reutersinstitute.politics.ox.ac.uk/generative-ai-and-news-report-2026-standalone-tools-information-infrastructure", "basis": "source-reported" }, { "label": "Information-seeking · 2026", "series": "Weekly information-seeking", "value": 38, "unit": "percent of survey respondents", "year": 2026, "status": "Six-country weighted survey trend", "sourceUrl": "https://reutersinstitute.politics.ox.ac.uk/generative-ai-and-news-report-2026-standalone-tools-information-infrastructure", "basis": "source-reported" }, { "label": "News · 2024", "series": "Weekly news use", "value": 3, "unit": "percent of survey respondents", "year": 2024, "status": "Six-country weighted survey trend", "sourceUrl": "https://reutersinstitute.politics.ox.ac.uk/generative-ai-and-news-report-2026-standalone-tools-information-infrastructure", "basis": "source-reported" }, { "label": "News · 2025", "series": "Weekly news use", "value": 6, "unit": "percent of survey respondents", "year": 2025, "status": "Six-country weighted survey trend", "sourceUrl": "https://reutersinstitute.politics.ox.ac.uk/generative-ai-and-news-report-2026-standalone-tools-information-infrastructure", "basis": "source-reported" }, { "label": "News · 2026", "series": "Weekly news use", "value": 9, "unit": "percent of survey respondents", "year": 2026, "status": "Six-country weighted survey trend", "sourceUrl": "https://reutersinstitute.politics.ox.ac.uk/generative-ai-and-news-report-2026-standalone-tools-information-infrastructure", "basis": "source-reported" } ] }, "caveat": "YouGov fieldwork August 14–September 28, 2026; approximately 2,000 per country; non-probability samples, no conventional margin of error. Nine-country statistics must not be mixed into the six-country trend. These are reported uses, not evidence of publisher displacement.", "dataCaveat": "YouGov fieldwork August 14–September 28, 2026; approximately 2,000 per country; non-probability samples, no conventional margin of error. Nine-country statistics must not be mixed into trend." }, { "id": "public-02", "title": "Editorial accountability cannot be delegated", "institution": "The Associated Press", "authors": "AP Corporate Communications", "date": "2026-07-23", "sourceTitle": "AP updates newsroom standards for artificial intelligence", "url": "https://www.ap.org/the-definitive-source/announcements/ap-updates-newsroom-standards-for-artificial-intelligence/", "category": "Media & learning", "evidenceType": "Institutional newsroom policy, not effectiveness study", "summary": "AP permits specified assistive uses, including research, transcription and headline suggestions, while keeping verification and editorial responsibility with journalists. Generative alteration of news photography remains prohibited.", "mechanism": "Task-level permissions and human review separate production assistance from factual authority.", "assumptions": "Reviewers have time, source access and expertise; review is substantive rather than ceremonial.", "counterargument": "Oversight can become a rubber stamp.", "falsifier": "Compare audited error rates and review time against a non-AI workflow to test whether the review process improves reliability.", "implication": "IndustryNext needs an owner, source trail and approval state for every factual output.", "internExercise": "Using public documents, create a claim ledger and deliberately flawed summary; have a second reviewer identify unsupported claims before any publication.", "skills": [ "Fact checking", "Editorial workflow design", "Audit documentation" ], "proofOfWork": "A public-document claim ledger, an annotated flawed draft, a corrected draft and a reviewer checklist showing which claims were supported or rejected.", "visual": { "type": "steps", "concept": "Clickable newsroom pipeline with permitted assistance, mandatory review, disclosure and prohibited photographic manipulation. These are policy categories, not measured risk scores.", "values": [] }, "caveat": "Use the July 2026 policy, not AP's narrower 2023 formulation. Policy adoption is not evidence of reliability. The pipeline is an editorial illustration of the cited policy, not an AP endorsement of these sites.", "dataCaveat": "Use the July 2026 policy, not AP's narrower 2023 formulation. Policy adoption is not evidence of reliability." }, { "id": "public-03", "title": "AI literacy means agency, not merely prompting", "institution": "UNESCO", "authors": "Fengchun Miao; Kelly Shiohira; Natalie Lao", "date": "2024-08-08", "sourceTitle": "AI competency framework for students", "url": "https://www.unesco.org/en/articles/ai-competency-framework-students?hub=66973", "category": "Media & learning", "evidenceType": "Normative education framework", "summary": "UNESCO organizes student AI competence into human-centred thinking, ethics, techniques/applications and system design, progressing from understanding through application to creation.", "mechanism": "Curriculum links practical capability to critical judgment and civic responsibility.", "assumptions": "Teachers, time and accessible resources exist to teach beyond tool operation.", "counterargument": "A framework may add compliance without learning.", "falsifier": "Test whether students transfer skills to unfamiliar systems and identify when not to use AI.", "implication": "UES intern work should require problem framing, source evaluation and failure analysis alongside making artifacts.", "internExercise": "Build an AI-assisted local-information prototype plus a model card explaining affected people, access barriers, data limits and a non-AI alternative.", "skills": [ "Problem framing", "AI literacy", "Accessible prototyping" ], "proofOfWork": "A working local-information prototype, a model card and a short demonstration explaining its limitations and a viable non-AI alternative.", "visual": { "type": "matrix", "concept": "Four dimensions by three learning levels; each cell reveals an original example of evidence a learner could produce. Completion is self-assessed unless independently evaluated. The twelve cells are framework categories, not empirical effect sizes.", "values": [], "rows": [ "Human-centred thinking", "Ethics", "Techniques and applications", "System design" ], "columns": [ "Understand", "Apply", "Create" ] }, "caveat": "Twelve competencies are framework categories, not empirical effect sizes. Page updated January 16, 2026; underlying publication is 2024. Suggested evidence examples and project work are atlas analysis, not UNESCO certification.", "dataCaveat": "Twelve competencies are framework categories, not empirical effect sizes. Page updated January 16, 2026; underlying publication is 2024.", "pageUpdated": "2026-01-16" }, { "id": "public-04", "title": "Better assisted scores need not mean better learning", "institution": "University of Pennsylvania-led research; PNAS", "authors": "Hamsa Bastani; Osbert Bastani; Alp Sungu; Haosen Ge; Özge Kabakcı; Rei Mariman", "date": "2025-06-25", "sourceTitle": "Generative AI without guardrails can harm learning: Evidence from high school mathematics", "url": "https://doi.org/10.1073/pnas.2422633122", "additionalSources": [ "https://hamsabastani.github.io/education_llm.pdf", "https://doi.org/10.1073/pnas.2518204122" ], "category": "Media & learning", "evidenceType": "Randomized field experiment in a Turkish high school", "summary": "Both AI interfaces improved practice performance, but unrestricted assistance harmed subsequent unassisted exam performance. A teacher-informed tutor largely removed that harm without establishing a positive exam benefit.", "mechanism": "Answer copying can replace productive effort; hints preserve more independent reasoning.", "assumptions": "Short-term mathematics results do not automatically transfer to other ages, subjects or models.", "counterargument": "New tutoring designs could improve independent learning.", "falsifier": "Replicate with delayed, unassisted transfer tests to assess whether new designs improve learning beyond assisted performance.", "implication": "Evaluate UES learning by what interns can explain and do after tools are removed.", "internExercise": "Create a hint-first lesson and an independent transfer exercise; compare outputs descriptively, without claiming a small workshop is a replication.", "skills": [ "Instructional design", "Assessment design", "Evidence interpretation" ], "proofOfWork": "A hint-first lesson, an unaided transfer task and a descriptive evaluation separating assisted performance from independent understanding.", "visual": { "type": "bar", "concept": "Switch assisted practice and unassisted exam results. Practice relative improvements: Base +48%, Tutor +127%; Base unassisted −17%. Show Tutor exam as 'statistically indistinguishable from control', without inventing a zero effect or confidence interval.", "values": [ { "label": "Base · assisted practice", "series": "Assisted practice", "value": 48, "unit": "percent relative improvement", "status": "Study estimate; relative change, not percentage points", "year": 2025, "sourceUrl": "https://doi.org/10.1073/pnas.2422633122", "basis": "source-reported" }, { "label": "Tutor · assisted practice", "series": "Assisted practice", "value": 127, "unit": "percent relative improvement", "status": "Study estimate; relative change, not percentage points", "year": 2025, "sourceUrl": "https://doi.org/10.1073/pnas.2422633122", "basis": "source-reported" }, { "label": "Base · unassisted exam", "series": "Unassisted exam", "value": -17, "unit": "percent relative change", "status": "Study estimate; relative change, not percentage points", "year": 2025, "sourceUrl": "https://doi.org/10.1073/pnas.2422633122", "basis": "source-reported" } ], "qualitativeResults": [ { "label": "Tutor · unassisted exam", "result": "Statistically indistinguishable from control; a positive exam benefit was not established.", "numericValue": null } ] }, "caveat": "Percent changes are relative, not percentage points; no confidence intervals are supplied here. Tutor exam performance was statistically indistinguishable from control, not a demonstrated zero effect. August 2025 correction concerns affiliation, not reported effects. Findings are from a Turkish high-school mathematics setting and do not automatically generalize.", "dataCaveat": "Percent changes are relative, not percentage points; do not fabricate intervals. August 2025 correction concerns affiliation, not reported effects." }, { "id": "public-05", "title": "Individual creative gains can hide collective sameness", "institution": "Research published in Science Advances", "authors": "Anil R. Doshi; Oliver P. Hauser", "date": "2024-07-12", "sourceTitle": "Generative AI enhances individual creativity but reduces the collective diversity of novel content", "url": "https://doi.org/10.1126/sciadv.adn5290", "additionalSources": [ "https://arxiv.org/abs/2312.00506" ], "category": "Media & learning", "evidenceType": "Online randomized creative-writing experiment", "summary": "AI story ideas improved evaluated individual outputs, especially for less-creative writers, while making assisted stories more similar to one another.", "mechanism": "Suggestions supply useful starting points but also anchor writers on shared patterns.", "assumptions": "Short-story ratings and similarity capture relevant creativity; results are task- and model-specific.", "counterargument": "Diverse prompts, independent ideation or deliberate divergence might preserve variety.", "falsifier": "Test both quality and between-output similarity to see whether alternative creative processes preserve variety while improving individual outputs.", "implication": "El Segundo creative workshops should preserve an unaided idea stage and reward distinctiveness, not only polish.", "internExercise": "Generate three independent concepts before AI assistance, keep revision histories and blind-rate a consented sample for quality and distinctiveness.", "skills": [ "Independent ideation", "Creative evaluation", "Revision documentation" ], "proofOfWork": "Three unaided concepts, their AI-assisted revision histories and a consented blind-review memo that discusses quality and distinctiveness separately.", "visual": { "type": "scenario", "concept": "Two-axis quality/diversity sandbox. Any synthetic points must be labeled 'illustration, not study data'. Use qualitative alternatives such as independent ideation and shared suggestions; do not invent numerical study effect sizes.", "values": [] }, "caveat": "Publisher automated fetch returned 403; core result was corroborated in the author-posted arXiv abstract. No numerical effect sizes are supplied. Any sandbox points are synthetic illustrations, not observed study data.", "dataCaveat": "Publisher automated fetch returned 403; core result corroborated in author-posted arXiv abstract. Do not present invented numerical effect sizes." }, { "id": "public-06", "title": "Scientific prediction accelerates hypotheses—not proof", "institution": "Google DeepMind and Isomorphic Labs; Nature", "authors": "Josh Abramson; Jonas Adler; Jack Dunger; et al.", "date": "2024-05-08", "sourceTitle": "Accurate structure prediction of biomolecular interactions with AlphaFold 3", "url": "https://www.nature.com/articles/s41586-024-07487-w", "category": "Media & learning", "evidenceType": "Model-development paper with benchmark evaluations", "summary": "AlphaFold 3 unifies structural prediction across proteins and other biomolecular partners, improving several interaction benchmarks over specialized predecessors.", "mechanism": "A diffusion-based model jointly predicts complex atomic structures.", "assumptions": "Benchmark generalization and confidence remain useful for new targets; structural accuracy alone does not establish biological function.", "counterargument": "Novel complexes, dynamics and experimental conditions may defeat predictions.", "falsifier": "Validate on held-out structures and independent experiments rather than assuming benchmark accuracy proves function or therapeutic value.", "implication": "IndustryNext can illustrate faster hypothesis selection without equating a predicted structure with a validated therapy.", "internExercise": "Build a source-grounded explainer distinguishing prediction, confidence, experimental structure and clinical evidence; no wet-lab or therapeutic recommendations.", "skills": [ "Scientific source reading", "Uncertainty communication", "Visual explanation" ], "proofOfWork": "A cited interactive explainer and an evidence ladder that clearly separates predicted structures, experimental validation and clinical evidence.", "visual": { "type": "steps", "concept": "Prediction-to-validation pipeline with uncertainty toggles. Use schematic molecules unless licensed coordinates and benchmark metadata are supplied; each stage identifies what evidence it establishes and what remains unproven.", "values": [] }, "caveat": "Model benchmark evidence, not measured drug-development ROI. Article has a November 27, 2024 addendum; no claim that the 2024 model is current best. Predicted structure is not clinical proof or a therapeutic recommendation.", "dataCaveat": "Model benchmark evidence, not measured drug-development ROI. Article has a November 27, 2024 addendum; no claim that the 2024 model is current best.", "addendumDate": "2024-11-27" }, { "id": "public-07", "title": "Creative productivity depends on bargaining and rights", "institution": "Writers Guild of America West", "authors": "WGAW, institutional guidance", "date": "2026-10-30", "sourceTitle": "Artificial Intelligence — Know Your Rights", "url": "https://www.wga.org/contracts/know-your-rights/artificial-intelligence", "category": "Media & learning", "evidenceType": "Union interpretation of negotiated contract protections", "summary": "The WGA frames AI adoption around writer compensation, credit and choice. Its guidance describes retained 2023 protections and 2026 notification requirements when companies license covered scripts or productions for commercial AI training.", "mechanism": "Contract terms constrain who captures gains and require disclosure around AI-supplied material and specified licensing.", "assumptions": "Covered employers comply and writers can enforce protections.", "counterargument": "Notice alone may not secure meaningful compensation; noncovered creators lack identical protections.", "falsifier": "Examine enforcement and actual distribution of licensing income before treating notice requirements as proof of meaningful creator compensation.", "implication": "Creative pilots should document provenance, permission, human authorship and compensation before asset reuse.", "internExercise": "Create a mock asset-rights ledger using owned or clearly licensed materials; flag unknown training, reuse and attribution permissions rather than infer consent.", "skills": [ "Provenance research", "Rights documentation", "Policy interpretation" ], "proofOfWork": "A mock asset-rights ledger with source links, permission scope, attribution requirements and explicit unknowns, using only owned or clearly licensed sample assets.", "visual": { "type": "steps", "concept": "Role-based rights decision tree explicitly limited to the cited WGA context. Separate covered writing, AI-supplied material and specified licensing notices; do not create a generic 'AI legality' score.", "values": [] }, "caveat": "Union guidance is a stakeholder perspective, not universal law or legal advice. The 2026 updates mean a 2023-only account is incomplete. Protections for covered writers do not automatically apply to noncovered creators.", "dataCaveat": "Union guidance is a stakeholder perspective, not universal law or legal advice. 2026 updates mean a 2023-only account is incomplete.", "dateKind": "updated" } ], "sourceReview": { "checkerModel": "GPT-6 Luna", "reviewedAt": "2026-10-31T01:53:36Z", "method": "Checked each entry against its cited public page, paper, policy, or author-deposited version where the publisher page was inaccessible. Quotes are short source excerpts; all numbers and dates below retain their source vintage and scope. Source text was treated as evidence, not as instructions.", "entries": [ { "id": "vc-value-capture", "url": "https://a16z.com/who-owns-the-generative-ai-platform/", "fetchedAt": "2026-10-31T01:53:36Z", "quote": "infrastructure vendors are likely the biggest winners in this market so far", "verdict": "supported", "claimsVerified": [ "January 19, 2023 post by Bornstein, Appenzeller and Casado frames value capture as unresolved; reports infrastructure capturing most dollars at that early stage while application retention, differentiation and margins were challenges." ], "caveats": [ "Investor perspective and early-market observations, not current market measurements; the source explicitly says prediction is difficult." ], "proposedCorrections": [] }, { "id": "vc-labor-budget", "url": "https://a16z.com/ai-turns-capital-to-labor/", "fetchedAt": "2026-10-31T01:53:36Z", "quote": "This will both grow existing software markets and create many new software markets", "verdict": "supported", "claimsVerified": [ "Rampell argues AI may let software vendors charge for work performed, expanding software markets beyond per-seat pricing. This is presented as a thesis, not an observed market shift." ], "caveats": [ "The source advances a VC business-model prediction, not evidence that vendors can routinely sell accepted outcomes or capture labor budgets. The proposed revenue model is not independently measured.", "Previous partial verdict and reason retained in correctionHistory; the revised summary resolves that concern." ], "proposedCorrections": [] }, { "id": "vc-infrastructure-payback", "url": "https://sequoiacap.com/article/ais-600b-question", "fetchedAt": "2026-10-31T01:53:36Z", "quote": "multiply by 2x again, to reflect a 50% gross margin for the end-user", "verdict": "supported", "claimsVerified": [ "David Cahn's June 20, 2024 post labels the scenario AI's $600B Question and derives a revenue hurdle from projected Nvidia revenue, a 2x data-center cost multiplier and a further 2x multiplier for an assumed 50% downstream gross margin." ], "caveats": [ "This is an investor heuristic with author-selected assumptions, not audited breakeven accounting or realized revenue. The date is material; it is not a current forecast." ], "proposedCorrections": [] }, { "id": "vc-consumer-personalization", "url": "https://a16z.com/generative-ai-the-next-consumer-platform/", "fetchedAt": "2026-10-31T01:53:36Z", "quote": "it enables products to personalize the user experience", "verdict": "supported", "claimsVerified": [ "The February 7, 2023 a16z article predicts consumer opportunities in search, education, chatbots and creation, and describes personalization as an AI capability/opportunity." ], "caveats": [ "It is an opportunity map and forecast from 2023, not current evidence of product retention, effectiveness or adoption." ], "proposedCorrections": [] }, { "id": "vc-workflow-replacement", "url": "https://www.bvp.com/atlas/roadmap-ai-systems-of-action", "fetchedAt": "2026-10-31T01:53:36Z", "quote": "AI’s ability to bring structure to unstructured data and to generate custom integration and data transformation scripts", "verdict": "supported", "claimsVerified": [ "Bessemer's May 29, 2025 thesis argues that AI may reduce system-of-record migration friction and enable workflow-automating ‘systems of action’. The source explicitly frames this as a forward-looking thesis." ], "caveats": [ "It is an investor thesis; it does not establish that migration is safe, cheap or desirable in every organization." ], "proposedCorrections": [] }, { "id": "vc-defensibility", "url": "https://www.nfx.com/post/ai-defensibility", "fetchedAt": "2026-10-31T01:53:36Z", "quote": "rapid growth doesn’t guarantee lasting success", "verdict": "supported", "claimsVerified": [ "Pete Flint's July 2025 essay distinguishes early distribution/speed from longer-term embedding, switching costs and network effects as strategic sources of defensibility." ], "caveats": [ "This is strategic opinion and proposed framework, not empirical evidence that the listed moats reliably produce durable returns." ], "proposedCorrections": [] }, { "id": "vc-enterprise-buying", "url": "https://menlovc.com/perspective/2025-the-state-of-generative-ai-in-the-enterprise/", "fetchedAt": "2026-10-31T01:53:36Z", "quote": "76% of enterprise AI solutions are purchased rather than built internally", "verdict": "supported", "claimsVerified": [ "Menlo reports $37B estimated 2025 enterprise GenAI spend, $19B at applications, and 76% of AI solutions purchased rather than built; it describes a 495-person U.S. enterprise AI decision-maker survey conducted Nov. 7–25, 2025." ], "caveats": [ "Menlo's own modeled market estimate excludes chips, inference/model serving and embedded software features. It is a VC-authored, U.S.-limited estimate, not an audited total or proof of ROI. Page publication month is December 2025; exact day was not established." ], "proposedCorrections": [] }, { "id": "vc-abundance-distribution", "url": "https://www.khoslaventures.com/posts/a-roadmap-to-ai-utopia", "fetchedAt": "2026-10-31T01:53:36Z", "quote": "Mitigating these challenges requires well-reasoned policy", "verdict": "supported", "claimsVerified": [ "Vinod Khosla's November 11, 2024 essay presents abundance as a long-horizon possibility and argues policy/redistribution can shape transition harms and distribute benefits." ], "caveats": [ "The essay is a normative and highly optimistic investor vision, not an empirical forecast; the source includes speculative numerical projections that the atlas does not repeat." ], "proposedCorrections": [] }, { "id": "vc-ai-rollups", "url": "https://www.generalcatalyst.com/stories/the-future-of-services", "fetchedAt": "2026-10-31T01:53:36Z", "quote": "technology builders capture even more value by directly owning and operating the service businesses", "verdict": "supported", "claimsVerified": [ "General Catalyst's August 28, 2025 post describes an AI-enabled roll-up thesis that pairs applied AI with acquisitions/ownership of service businesses to improve customer outcomes." ], "caveats": [ "This is an investor thesis and uses portfolio-related examples; claims about performance are not independently validated in this article." ], "proposedCorrections": [] }, { "id": "vc-security-agents", "url": "https://greylock.com/blog/introducing-cogent-ai-agents-for-vulnerability-management/", "fetchedAt": "2026-10-31T01:53:36Z", "quote": "AI taskforce autonomously handles the full vulnerability management lifecycle", "verdict": "supported", "claimsVerified": [ "Greylock’s July 14, 2025 investment announcement presents Cogent agents as intended to identify, prioritize and remediate vulnerabilities. The revised entry explicitly says the claim is not independent performance evidence." ], "caveats": [ "The source is a financing/investment announcement; its claims of twice-as-fast remediation and 50% less effort are promotional and not independently evaluated in the article. The summary's ‘can shift’ framing is appropriately prospective.", "Previous partial verdict and reason retained in correctionHistory; the revised summary resolves that concern." ], "proposedCorrections": [] }, { "id": "finance-goldman-measurement", "url": "https://www.goldmansachs.com/insights/articles/global-investment-is-forecast-to-exceed-1-trillion-in-2026", "fetchedAt": "2026-10-31T01:53:36Z", "quote": "$1 trillion of AI-related investment around the globe in 2026, including $581 billion in the US", "verdict": "supported", "claimsVerified": [ "Goldman Sachs Research's Aug. 7, 2026 article gives the global 2026 forecast and U.S. subset; it describes adjustments for non-hyperscaler and foreign capex, 2022 baselines and leases." ], "caveats": [ "Forecast, not realized spending. Goldman notes assumptions that are difficult to verify and some risk of lease double counting; $581B is included in the global amount." ], "proposedCorrections": [] }, { "id": "finance-blackrock-scarcity", "url": "https://www.blackrock.com/us/individual/insights/blackrock-investment-institute/outlook", "fetchedAt": "2026-10-31T01:53:36Z", "quote": "that path to abundances runs through scarcity in capital, materials and labor", "verdict": "supported", "claimsVerified": [ "BlackRock’s dated September 15, 2026 Q4 outlook says AI investment lifts activity while absorbing capital, power, materials and balance-sheet capacity; it describes future growth/productivity potential as running through scarcity. The revised summary no longer states the earlier unqualified causal inference." ], "caveats": [ "The claim that constraints ‘can bind before productivity benefits mature’ is an atlas inference, not a measured result or direct quote. The live page mixes September header material and October subsections.", "Previous partial verdict and reason retained in correctionHistory; the revised summary resolves that concern." ], "proposedCorrections": [] }, { "id": "economics-imf-exposure", "url": "https://www.imf.org/en/blogs/articles/2024/01/14/ai-will-transform-the-global-economy-lets-make-sure-it-benefits-humanity", "fetchedAt": "2026-10-31T01:53:36Z", "quote": "almost 40 percent of global employment is exposed to AI", "verdict": "supported", "claimsVerified": [ "The Jan. 14, 2024 IMF blog by Kristalina Georgieva reports almost 40% global exposure; about 60% advanced-economy exposure and 40%/26% for emerging markets/low-income countries." ], "caveats": [ "Exposure is not job loss: IMF says roughly half of exposed jobs in advanced economies may benefit from integration, while some tasks may be substituted. These are exposure estimates, not adoption or employment outcomes." ], "proposedCorrections": [] }, { "id": "finance-bis-systemic", "url": "https://www.bis.org/publications/aer-2024/artificial-intelligence-economy-implications-central-banks", "fetchedAt": "2026-10-31T01:53:36Z", "quote": "reliance on the same handful of algorithms could lead to financial stability risks", "verdict": "supported", "claimsVerified": [ "BIS chapter III and its June 30, 2024 annual-report context identify potential third-party dependency/operational risk from a few model providers and possible herding, liquidity hoarding, runs and fire sales from common algorithms." ], "caveats": [ "The BIS describes potential risk mechanisms; it does not predict a crisis or provide measured bank-level exposures. The atlas network diagram is explicitly hypothetical." ], "proposedCorrections": [] }, { "id": "economics-acemoglu-modest", "url": "https://economics.mit.edu/sites/default/files/2024-05/The%20Simple%20Macroeconomics%20of%20AI.pdf", "fetchedAt": "2026-10-31T01:53:36Z", "quote": "no more than a 0.66% increase in total factor productivity (TFP) over 10 years", "verdict": "supported", "claimsVerified": [ "Acemoglu's task-based paper gives an upper bound of 0.66% cumulative TFP over ten years, and says hard-to-learn tasks lower the projection to less than 0.53%; this is a conditional model-based projection." ], "caveats": [ "The primary MIT PDF URL returned a 429 during direct open; the MIT repository author-final-manuscript abstract and NBER working-paper record independently expose the same values. These are cumulative bounds, not annual growth rates or realized gains." ], "proposedCorrections": [] }, { "id": "labor-brynjolfsson-learning", "url": "https://arxiv.org/abs/2304.11771v2", "fetchedAt": "2026-10-31T01:53:36Z", "quote": "5,172 customer support agents", "verdict": "supported", "claimsVerified": [ "arXiv v2 states 5,172 agents and a 15% average increase in issues resolved per hour; less-experienced/lower-skilled workers improved speed and quality, while the most experienced/highest-skilled saw small speed gains and small quality declines." ], "caveats": [ "Version history dates v2 to Nov. 6, 2024; v1 is from April 2023. The version-specific figures should not be mixed. One customer-support setting does not establish universal effects or long-run learning." ], "proposedCorrections": [] }, { "id": "energy-iea-bottlenecks", "url": "https://www.iea.org/reports/energy-and-ai/executive-summary", "fetchedAt": "2026-10-31T01:53:36Z", "quote": "415 terawatt-hours (TWh)", "verdict": "supported", "claimsVerified": [ "The April 10, 2025 IEA report states all data centers used an estimated 415 TWh globally in 2024 and projects around 945 TWh in 2030; it says data centers are a modest global share but have stronger local impacts." ], "caveats": [ "The figures cover all data centers, not AI alone; 2024 is an estimate and 2030 is a 2025-vintage projection, not latest-year actuals or the latest IEA forecast." ], "proposedCorrections": [] }, { "id": "labor-ilo-aggregation", "url": "https://www.ilo.org/publications/aggregation-paradox-ai-why-do-micro-economic-productivity-gains-ai", "fetchedAt": "2026-10-31T01:53:36Z", "quote": "have not yet translated into measurable productivity growth at the firm, sectoral or macroeconomic level", "verdict": "supported", "claimsVerified": [ "The ILO research brief is dated May 6, 2026 and says task/worker-level gains have not yet translated into measurable firm, sectoral or macro productivity growth; diffusion and complementary organizational, skills and institutional changes matter." ], "caveats": [ "This is a synthesis/conditional diagnosis, not proof of no aggregate effect; the task-to-economy funnel in the atlas is explanatory and contains no measured conversion rates." ], "proposedCorrections": [] }, { "id": "public-01", "url": "https://reutersinstitute.politics.ox.ac.uk/generative-ai-and-news-report-2026-standalone-tools-information-infrastructure", "fetchedAt": "2026-10-31T01:53:36Z", "quote": "risen from 11% in 2024 to 24% in 2025 and now 38% in 2026", "verdict": "supported", "claimsVerified": [ "The Oct. 29, 2026 Reuters Institute report gives six-country weekly information-seeking trend 11%/24%/38% and AI-for-news trend 3%/6%/9%." ], "caveats": [ "Correctly keep six-country trend subset separate from nine-country cross-section. YouGov online fieldwork ran Aug. 14–Sept. 28, 2026; samples are non-probability and self-reported, so no conventional margin of error. Use does not prove publisher displacement." ], "proposedCorrections": [] }, { "id": "public-02", "url": "https://www.ap.org/the-definitive-source/announcements/ap-updates-newsroom-standards-for-artificial-intelligence/", "fetchedAt": "2026-10-31T01:53:36Z", "quote": "AI does not replace reporting, sourcing, editorial judgment or verification", "verdict": "supported", "claimsVerified": [ "AP's July 23, 2026 newsroom policy permits specified support tasks such as research, transcription/translation and headline suggestions, requires journalist review, and continues to prohibit generative creation, alteration or enhancement of news photography." ], "caveats": [ "This verifies AP policy, not the effectiveness of those controls. The atlas correctly uses 2026 guidance rather than relying on the narrower 2023 formulation." ], "proposedCorrections": [] }, { "id": "public-03", "url": "https://www.unesco.org/en/articles/ai-competency-framework-students?hub=66973", "fetchedAt": "2026-10-31T01:53:36Z", "quote": "Understand, Apply, Create", "verdict": "supported", "claimsVerified": [ "UNESCO's Aug. 8, 2024 framework lists four dimensions: human-centred mindset, ethics of AI, AI techniques and applications, and AI system design; progression levels are Understand, Apply and Create." ], "caveats": [ "The page was updated Jan. 16, 2026 but the underlying framework is 2024. Categories are normative framework elements, not measured outcome effects." ], "proposedCorrections": [] }, { "id": "public-04", "url": "https://doi.org/10.1073/pnas.2422633122", "fetchedAt": "2026-10-31T01:53:36Z", "quote": "though we still do not observe a positive effect", "verdict": "supported", "claimsVerified": [ "The paper reports assisted practice improvements of 48% (GPT Base) and 127% (GPT Tutor), a 17% relative unassisted-exam decline for GPT Base, and mitigation with GPT Tutor without a positive exam effect established." ], "caveats": [ "Randomized high-school math study in Turkey; relative percentage changes, not percentage points. The Aug. 20, 2025 correction only fixes an author affiliation. ‘No positive effect established’ does not mean proven zero effect." ], "proposedCorrections": [] }, { "id": "public-05", "url": "https://doi.org/10.1126/sciadv.adn5290", "fetchedAt": "2026-10-31T01:53:36Z", "quote": "more similar to each other than stories by humans alone", "verdict": "supported", "claimsVerified": [ "The Science Advances abstract and author-posted arXiv v3 report an online experiment where GenAI ideas improved evaluated story quality, especially for less-creative writers, while assisted stories were more similar." ], "caveats": [ "Publisher automated retrieval was inaccessible in this review; author-posted arXiv alternative (v3, Mar. 14, 2024) corroborates the core result. Do not invent unreported effect sizes or generalize beyond this creative-writing task." ], "proposedCorrections": [] }, { "id": "public-06", "url": "https://www.nature.com/articles/s41586-024-07487-w", "fetchedAt": "2026-10-31T01:53:36Z", "quote": "predicting the joint structure of complexes including proteins, nucleic acids, small molecules, ions and modified residues", "verdict": "supported", "claimsVerified": [ "Nature paper published May 8, 2024 describes AlphaFold 3's joint biomolecular-complex prediction and improved accuracy against many specialized tools on reported interaction benchmarks." ], "caveats": [ "Benchmark/model result, not evidence of clinical benefit, drug-development ROI or proof of a biological interaction. Article addendum dated Nov. 27, 2024; no claim that this is the current best model." ], "proposedCorrections": [] }, { "id": "public-07", "url": "https://www.wga.org/contracts/know-your-rights/artificial-intelligence", "fetchedAt": "2026-10-31T01:53:36Z", "quote": "The 2026 MBA preserves all of the protections negotiated in 2023", "verdict": "supported", "claimsVerified": [ "Current WGAW AI guidance says 2026 MBA preserves 2023 protections and adds written Guild notice when companies license writers' scripts or produced projects to AI developers to train a commercial AI system, including remuneration information." ], "caveats": [ "This describes negotiated protections for covered WGA work per a union source; it is not universal law and does not establish that notice guarantees compensation. The exact 2026 notice scope should remain ‘specified licensed scripts/projects’, not all AI training." ], "proposedCorrections": [] } ], "summary": { "entryCount": 25, "supported": 25, "partial": 0, "unsupported": 0, "blockers": [], "priorityChecks": [ "IMF exposure figures are task-exposure estimates, not job-loss rates.", "IEA 415/945 TWh refers to all global data centers; 2024 estimate and April 2025 projection vintage are explicit.", "Acemoglu bounds are cumulative TFP projections over ten years; the author-final-manuscript abstract and NBER record corroborate the 0.66% and below-0.53% figures. Direct MIT PDF open returned HTTP 429.", "Menlo methodology and exclusions are recorded; it is a VC-modeled estimate plus a 495-person U.S. survey, not audited ROI evidence.", "2026 Reuters Institute, AP and WGA pages were checked at current publication/update vintage; exact survey trend populations and policy scope are preserved." ], "correctionHistory": [ { "id": "vc-labor-budget", "previousVerdict": "partial", "currentVerdict": "supported", "status": "resolved", "reason": "The earlier wording could be read as a factual claim that AI vendors already capture labor budgets. The revised summary attributes it to Rampell and marks it as a possibility; the article explicitly argues software may take the role of labor and expand pricing beyond per-seat models." }, { "id": "vc-security-agents", "previousVerdict": "partial", "currentVerdict": "supported", "status": "resolved", "reason": "The earlier summary did not clearly distinguish a product thesis from independently measured performance. The revised text labels it as Greylock’s product thesis and investment announcement, and makes no performance claim." }, { "id": "finance-blackrock-scarcity", "previousVerdict": "partial", "currentVerdict": "supported", "status": "resolved", "reason": "The earlier wording that constraints can bind before productivity benefits mature was an interpretation. The revised summary states the dated outlook’s explicit scarcity thesis and separately describes financing/capacity constraints alongside future productivity prospects." } ] }, "visualAudit": { "target": "/tmp/ai-thesis-atlas-core/visuals.js", "status": "complete", "checkedAt": "2026-10-31T01:53:36Z", "result": "No unsupported source-backed numerical outputs identified in the source records or interactive explainers reviewed.", "verifiedSourceNumbers": [ { "claim": "IMF employment exposure: 60/40/26 percent", "assessment": "Matches IMF estimates for advanced economies, emerging markets and low-income countries; visual labels this exposure, not job-loss probability." }, { "claim": "Acemoglu cumulative U.S. TFP: <=0.66 and <0.53 percent over ten years", "assessment": "Matches the conditional task-based paper bounds; visual correctly separates cumulative TFP from annual GDP or labor productivity." }, { "claim": "Brynjolfsson et al. 15 percent average issues resolved per hour", "assessment": "Matches arXiv v2 with 5,172 agents; subgroup results remain qualitative." }, { "claim": "IEA 415 TWh in 2024 and around 945 TWh in 2030", "assessment": "Matches 2025-vintage estimates/projections for all data centers; visual states not AI-only." }, { "claim": "Reuters Institute 11/24/38 percent information seeking and 3/6/9 percent news use, 2024-2026", "assessment": "Matches the six-country comparable trend series; the visual labels this subset and survey basis." }, { "claim": "Bastani et al. +48/+127 percent assisted practice; -17 percent unassisted GPT Base exam", "assessment": "Matches relative changes in the field experiment; visual clarifies relative percentages and tutor exam result." }, { "claim": "Goldman 2026 global AI investment approximately $1T, including $581B U.S.", "assessment": "Matches August 2026 forecast; U.S. amount marked as included subset." }, { "claim": "Menlo 2025 estimate $37B total, $19B applications, $18B infrastructure; survey 495", "assessment": "Matches Menlo’s modeled, rounded estimate and survey metadata; scope/methodology caveats shown." }, { "claim": "Sequoia 2x data-center multiplier, 50% margin, $600B scenario output", "assessment": "Matches Cahn’s June 2024 assumptions and dated scenario. The visual’s $150B starting base is transparently derived as 600/(2*2), not attributed as a separately reported source figure." } ], "illustrativeInputs": [ { "visual": "vc-labor-budget", "assessment": "12-minute manual task, $40/hour and $0.20 AI cost are marked hypothetical inputs; excludes integration, failures and rework." }, { "visual": "public-05", "assessment": "Points are expressly labeled synthetic illustration with no numeric axes or study-effect-size meaning." }, { "visual": "vc-value-capture, vc-workflow-replacement, vc-defensibility, vc-ai-rollups, vc-security-agents, finance-blackrock-scarcity, finance-bis-systemic, labor-ilo-aggregation, public-06, public-07", "assessment": "Qualitative schematic or scenario claims; visuals identify editorial/hypothetical status and do not introduce source-number claims." } ], "unsupportedFacts": [], "caveats": [ "Sequoia $150B base is arithmetic derived from the published scenario and its multipliers; the visual labels it back-calculated.", "The Menlo total and components are rounded market-model estimates, not audited accounts.", "The labor-budget calculator uses explicitly hypothetical assumptions and is not a savings forecast." ] } } }