FAQ
Frequently Asked Questions
Common questions about AI progress, our methodology, and how Future Shock works.
What is Future Shock?
Future Shock is an independent observatory that tracks the acceleration of artificial intelligence. We ingest from 40+ sources daily, run editorial review, and publish what actually matters across six axes: capability, compute, robotics, research, open source, and prediction accuracy. Data coverage starts at GPT-4's release in March 2023.
What is the AI takeoff?
AI takeoff refers to the period when AI capabilities begin improving at a self-reinforcing, accelerating rate — where AI progress begins to compound on itself. Future Shock tracks the early stages of this by monitoring benchmark progress, compute scaling, research velocity, and the convergence of expert AGI timelines.
How fast is AI actually improving?
Faster than most people realize. Since GPT-4 in March 2023, we've tracked over 2,300 significant events — benchmark records broken, compute milestones, robotics deployments, research breakthroughs. Context windows have grown from 4K tokens to multi-million token windows. Frontier models now score above 90% on exams that stumped earlier systems. The rate of improvement is accelerating, not flattening.
What is AGI, and when will it arrive?
Artificial General Intelligence (AGI) is loosely defined as AI that can perform any intellectual task a human can. There is no universally agreed definition — different researchers and labs use different framings. Our Predictions Tracker monitors public predictions from named experts with specific timeframes. As of early 2026, expert median estimates have converged toward the mid-2020s to early 2030s range, with significant variance. We do not make our own AGI predictions; we track and verify others'.
What benchmarks does Future Shock track?
We track the leading capability benchmarks: MMLU (general knowledge), HumanEval and SWE-bench (coding), MATH and AIME (mathematics), GPQA (graduate-level science), HLE (Humanity's Last Exam — the current frontier benchmark), and ARC-AGI (abstract reasoning). We also track benchmark saturation — when a benchmark is effectively 'solved' and no longer meaningful for differentiating models.
What is benchmark saturation?
Benchmark saturation occurs when AI systems score so high on a benchmark that it no longer distinguishes capability levels. MMLU is effectively saturated — top models all score above 87%. When a benchmark saturates, the field moves to harder ones. We track this progression to understand where the frontier actually is.
How does the Future Shock newsletter work?
We publish several newsletter editions: The Signal (daily top developments), The Long View (Sunday deep analysis), The Noise (Friday lighter roundup), Bright Signals (Tuesday: positive AI developments), Ops & Bloopers (Wednesday: behind-the-scenes of running an AI newsroom), and What-If (Thursday: counterfactual analysis). All are free at news.future-shock.ai.
Is Future Shock affiliated with any AI company?
No. Future Shock is independent. We use AI tools (including Anthropic's Claude for ingestion and processing), but we are not funded by, affiliated with, or editorially influenced by any AI company. We apply the same editorial standards to coverage of OpenAI, Anthropic, Google, Meta, and others.
Does Future Shock give financial advice?
Absolutely not. We cover AI progress — not stocks, not crypto, not investment strategy. Nothing on Future Shock should be construed as financial advice. We don't cover prediction markets or token prices.
How does the Predictions Tracker work?
We collect public predictions from named individuals — researchers, lab leaders, journalists, independent forecasters — where they make a specific, falsifiable claim with a timeframe. We track confidence assignments, resolve predictions when their target date passes, and publish accuracy scores. Details at future-shock.ai/predictions/methodology.
What is the Precondition Density Model (PDM)?
The PDM is an original analytical framework for evaluating how close AI is to achieving specific capabilities. Rather than asking 'will it happen?', PDM asks 'what needs to be true first?' — mapping preconditions (technical, data, infrastructure, social) and tracking how many are met. Read the full paper at future-shock.ai/research/precondition-density-model.
How do I subscribe to the newsletter?
Visit news.future-shock.ai and enter your email. All newsletters are free. You can choose which editions to receive.
How can I submit a correction or tip?
Email nic@future-shock.ai. We take accuracy seriously and correct errors promptly. If we've missed a significant AI development worth tracking, tips are welcome.
Don't see your question here? Email us or check the methodology page for details on how we track predictions.