2023-03 → Present
The AI takeoff is happening.
We're measuring it.
Future Shock maps the accelerating trajectory of artificial intelligence — across models, compute, robotics, research, and open source. Updated daily. No hype. No financial advice. Just the signal.
Six Tracking Axes
The dimensions of accelerationModel Capability
Benchmarks, multimodal, agents, context windows, tool use
Compute & Chips
GPU releases, FLOPs/dollar, training clusters, energy
Robotics
Humanoid milestones, autonomous deployments, LLM-driven demos
Research Velocity
arXiv publication rate, breakthrough papers, citation velocity
Open Source
GitHub commit velocity, AI co-authored code, framework growth
Policy
Government action, regulation, defense, trade, AI governance
Latest Signals
Updated dailyResearch: over 53M people in China worked as food delivery or ridesharing drivers as of 2025, up 10M in two years, amid accelerating automation in manufacturing (Financial Times)
Financial Times : Research: over 53M people in China worked as food delivery or ridesharing drivers as of 2025, up 10M in two years, amid accelerating automation in manufacturing — As construction slumps and manufacturing automates, 53mn Chinese now work for low wages in food delivery and ride-hailing.
AI4AI at Test-Time: Strong-to-Weak Capability Transfer via Harnesses
Recent work on distillation transfers the capabilities of large models to smaller ones often by updating the latter's parameters, through teacher forcing, on-policy distillation, and related training-time methods. In this paper, we ask whether such transfer can instead occur at test time. We study s...
AVA-Encoder: Towards Agent-Native Video Representation Learning
Creative agents still lack an effective way to learn from high-quality human films, limiting their ability to produce cinematic-grade videos. A key challenge is the absence of a structured video representation that is both faithful to film content and directly usable for agentic reasoning and manipu...
Beyond Trial-and-Error: Agentic Optimization for Image-to-Video Adherence
Modern black-box Image-to-Video (I2V) models offer powerful capabilities in automated content creation, yet their lack of fine-grained control and reliability presents significant challenges in professional workflows. Their inherent stochasticity causes minor variations in textual prompts or hyperpa...
Convergent Detour Hijacking: Task-Preserving Resource Amplification in Skill-Based LLM Agents
LLM agents increasingly rely on third-party skills, using natural-language descriptions for selection and instruction bodies for planning. This progressive-disclosure design exposes two sequential control points to untrusted publishers: a static skill may steer an otherwise correct task onto an unne...
Diagram-MMU: A Multi-Modal Benchmark for Scientific Diagrams
Multimodal Large Language Models (MLLMs) have been growing the capability for scientific writing and collaboration. For example, OpenAI Prism is a free workspace for scientific writing and collaboration. One important feature in Prism is turning scientific diagrams directly into LaTeX TikZ code. In ...
Featured Essays
Durable Future Shock pieces from the newsletter archive: agent coordination, AI infrastructure, security, and the systems around the models.
Work
The K-Curve
The widening gap between people who compound value from AI and those who tried it once and bounced.
Open source
The Quantization Underground
How a four-line experiment changed the way small AI models are compressed and run locally.
Coordination
We Made AI Agents Negotiate the End of the World
The same model behaved differently depending on which other agents were in the room.
Agents
The (Missing) Layer Between You and Your Agents
What is missing between the agents doing the work and the person trying to understand what happened.
Infrastructure
The Grid Under the Cloud
The next decade of AI will be shaped by the places that can power, cool, permit, and physically host it.
Accountability
Promise Nothing, Claim Everything
Six competing AI companies arrived at remarkably similar terms for warranties, liability, and training data.
Join early readers tracking AI's acceleration
The briefing that cuts through the noise
AI news, analysis, and weekly deep dives. No hype, no spam. (We read the papers so you don't have to.)
Powered by Ghost · Unsubscribe anytime