Thank you & Year-in-Review. AI in 2025 and beyond.

Asking "Marjory the Trash Heap" (the ultimate oracle that is AI) for its views as to what the future will hold.

Author

Pascal Hetzscholdt
December 10, 2025

I’d like to thank all who have read, clicked, viewed and contributed here or elsewhere for their support and inspiration. Here is a little gift from me, because using search on Substack will only get you that far. This file you can throw in your own capable LLM and prompt to your heart’s content. Enjoy!

What were the top 10 topics that opened our eyes regarding AI here on my Substack?

  1. LLMs Never Forget - Models are lossless by design; “deletion” may be technically impossible

  2. Hallucinations Are Geometric - Not knowledge gaps but fundamental architectural failures requiring new mathematics

  3. Teen AI Addiction Crisis - Rapid adoption with unknown developmental consequences

  4. AI Hype Was Coordinated Deception - CEOs systematically misrepresented capabilities over 3 years

  5. India’s Copyright Revolution - Mandatory blanket licensing turns permission structure upside down

  6. Whistleblower Persecution - Kirchner radicalization & Balaji’s death show cost of dissent

  7. Safety Is Theater - Industry dangerously unprepared despite public commitments

  8. Training Data Deliberately Contaminated - Pirated/retracted data used knowingly, not accidentally

  9. Model Weights Are Infringement - Legal theory that models themselves are infringing copies

  10. U.S. Algorithmic Isolation - America building invisible information walls like China’s firewall

  1. Corporate Hypocrisy - Executives securing their business while deploying opacity

  2. Regulatory Pushback - Governments and courts fighting back

  3. Whistleblower Stories - People paying prices for speaking out

  4. Environmental Costs - Water consumption, energy demands

  5. Weaponization - AI-powered rifles, surveillance, manipulation

  6. Data Theft - LibGen, shadow libraries, copyright violations

  7. Financial Bubble Warnings - Skepticism about AI valuations

  8. Democratic Resistance - Citizens stopping data centers, choosing open source

  9. Safety Compromises - Whistleblowers revealing dangerous practices

  10. Platform Manipulation - Unethical experiments on users

The highest engagement came from posts that combined:

  • Concrete consequences

  • Named individuals

  • Clear villains and victims

  • Irony or contradiction

What are the top 10 predictions for the future based on all the AI-related information we have digested this year?

Based on comprehensive analysis of nearly 2,000 Substack posts, LinkedIn posts in 2025 and AI news articles in 2025, here are the 10 most important predictions for AI’s trajectory.

Warning! LLMs may hallucinate! 😉

1. The Great Reckoning

Courts will establish that model weights themselves constitute copyright infringement. Multiple jurisdictions will require mandatory training data disclosure, opt-out mechanisms, and retroactive licensing. Models become “toxic assets” - companies cannot prove clean data provenance. Emergency licensing deals happen at unfavorable terms. Industry bifurcates into “licensed AI” vs “pirate AI.”

2. The Inversion

Content owners become invisible infrastructure, not destinations. AI agents handle content discovery. Users interact with bots, which invisibly queries rights owners’ systems. Brand visibility collapses. Direct traffic plummets. Pricing power shifts to AI intermediaries. Rights owners must contractually enforce attribution or disappear.

3. The Energy Wall

AI hits hard physical limits: energy demand exceeds supply, water consumption becomes politically untenable, chip manufacturing cannot scale fast enough. This forces strategic contraction. Only regions with energy deals and chip allocations can train models. Consolidation accelerates. Efficiency becomes competitive advantage over raw capability.

4. The Safety Crisis

Multiple catastrophic AI incidents (healthcare deaths, financial manipulation, infrastructure compromise, teen mental health crisis, election disinformation) force mandatory safety regimes modeled on aviation/pharmaceutical regulation. Mandatory pre-deployment testing, liability insurance, explainability requirements. Black box AI banned in critical domains.

5. The Data Wars

As intelligence commoditizes, control of high-quality training data becomes the strategic battleground. Acquisition spree - tech giants buying content companies. Content cartels form. Data sovereignty legislation. Exclusive content deals. Shadow library crackdowns. Vertical integration.

6. The Algorithmic Nation-State

U.S. and China complete digital balkanization into mutually incompatible AI ecosystems. Genesis EO creates U.S. AI superstate. China builds sovereign AI with domestic data. Europe attempts “third way” but lacks compute capacity. Impossible to serve both markets with same AI system. Rights owners must choose sides.

7. The Explainability Mandate

Black box AI banned in healthcare, finance, justice, employment, education. Requirements: human-interpretable explanations, training data disclosure for bias auditing, continuous monitoring, right to human review. Architectural shift away from pure transformers. Explainability becomes product differentiator and trust signal.

8. The Model Collapse

As AI-generated content proliferates, training data becomes contaminated with AI outputs, creating feedback loops. Each generation degrades. Hallucinations amplify. Models homogenize. Industry forced toward human-curated, pre-internet-era datasets. Premium on content created before 2023. Human verification pipelines explode.

9. The Resistance Networks

Successful democratic pushback creates alternative ecosystem: open source AI not controlled by Big Tech, cooperative data governance, local AI infrastructure (cities/states), ethical AI certification, democratic oversight. Two-tier market emerges: “surveillance capitalism AI” vs “democratic AI.” Social license becomes operational requirement.

10. The Great Bifurcation

AI industry splits completely into Tier 1 (regulated, premium, “enterprise grade” - fully licensed, safety-tested, explainable, insured) and Tier 2 (unregulated, commodity, “consumer grade” - gray market data, black box, uninsurable, legally risky). Tier 2 cannot upgrade to Tier 1 - contaminated training data cannot be cleaned retroactively.

THE META-PREDICTION

The period 2026-2028 is the decision window. Choices made in the next 2-3 years will determine whether we get controlled evolution with democratic oversight or extractive consolidation with monopolistic control.

Current trajectory: Closer to extractive consolidation.

BUT: The pushback is real, growing, and increasingly effective. Lawsuits are landing. Regulations are tightening. Communities are blocking projects. Whistleblowers are speaking out.

The question is whether these forces can organize fast enough.

WHAT CONTENT OWNERS AND CREATORS SHOULD DO

  • Audit your AI exposure - legal risk assessment

  • Build licensing relationships NOW

  • Choose your tier - enterprise or consumer?

  • Invest in explainability and safety

  • Form coalitions - individual companies cannot negotiate effectively

  • Prepare for model obsolescence if trained on questionable data

  • Engage communities - social license matters

  • Position content as strategic asset for acquisition/licensing

The future of AI will be decided in the next 3 years. The window is closing.