If you're a policy maker, organizational planner, or decision maker:
This page is particularly for you. The timeline below is not a forecast. It's a map of what has already happened and what researchers who study this professionally believe is coming. The goal isn't to alarm. It's to give you enough grounding in the actual evidence to make your own assessment of what your planning horizon should assume.
Some widely-reported milestones.
Each year, AI capabilities that experts predicted were decades away arrive months later. These are widely-reported milestones. Whether they imply what some commentators claim is a separate question.
AlexNet wins ImageNet by a record margin
Deep learning goes from academic curiosity to dominant paradigm overnight. Every major tech lab pivots to neural networks.
Krizhevsky, Sutskever, Hinton - NeurIPS 2012AlphaGo defeats world champion Lee Sedol 4-1
Go was considered a decade away from being solved. The win signals that pattern recognition at superhuman level is here.
DeepMind / NatureTransformer architecture published at Google
"Attention Is All You Need" replaces recurrent networks. Every major model since - GPT, BERT, Claude, Gemini - runs on this foundation.
Vaswani et al. - NeurIPS 2017GPT-2 withheld for being "too dangerous to release"
OpenAI self-censors a text model. Within 18 months, models 100x more capable are freely available.
OpenAIGPT-3: few-shot learning at 175B parameters
Researchers discover emergent abilities not present in smaller models. The scaling hypothesis becomes mainstream.
Brown et al. - OpenAIAlphaFold 2 solves the protein folding problem
50 years of unsolved biology in one model. DeepMind releases structures for nearly every known protein.
Jumper et al. - Nature 2021Stable Diffusion, DALL-E 2, and Midjourney go public
Image generation shifts from research demo to consumer product in months. Visual creative work is never the same.
Stability AI / OpenAI / MidjourneyChatGPT reaches 100 million users in 2 months
Fastest consumer product adoption in history. Faster than TikTok (9 months), Instagram (2.5 years), Netflix (10 years).
Reuters / UBSGPT-4 passes bar exam, medical boards, and CPA exam
AI scores in the 90th percentile of human professionals across multiple licensed fields simultaneously.
OpenAI Technical ReportGeoffrey Hinton leaves Google and warns of existential risk
One of the three "godfathers of deep learning" publicly says he regrets his life's work and fears loss of human control.
BBC NewsOpenAI o1 introduces chain-of-thought reasoning at scale
Models that "think before answering" beat PhD-level humans on physics and chemistry benchmarks.
OpenAISora generates minute-long photorealistic video from text
Video generation reaches cinematic quality. Hollywood studios begin negotiating with AI companies.
OpenAIClaude 3 Opus and Gemini Ultra match or exceed GPT-4
Multiple frontier models from competing labs. The AI race goes from one-horse to multi-horse in a single year.
Anthropic / GoogleClaude 3.7 Sonnet demonstrates extended autonomous reasoning
Models hold coherent thought across thousands of steps. Multi-hour agentic tasks become routine.
AnthropicAI coding agents write and ship production software
GitHub Copilot, Cursor, and Claude Code handle entire features end-to-end. Software engineers' role shifts to review and direction.
GitHub / AnthropicHumanoid robots begin factory deployment at scale
Figure, Tesla Optimus, and 1X begin measured production runs. The physical world becomes the next AI frontier.
Figure AI / Tesla / 1X TechnologiesProjected: AI outperforms experts in most white-collar tasks
PROJECTEDEconomic displacement begins in knowledge work. Legal, financial analysis, and radiology see major workforce restructuring.
McKinsey Global InstituteProjected: AI matches median human performance broadly
PROJECTEDLeopold Aschenbrenner's "Situational Awareness" timeline. The window for building independent skills and infrastructure closes.
situational-awareness.aiProjected: Autonomous AI research accelerates discovery
PROJECTEDAI systems begin meaningfully contributing to their own improvement. Human-paced science is no longer the bottleneck.
Metaculus community forecastProjected: Artificial General Intelligence
PROJECTEDMedian forecast from Metaculus aggregation (~2030). Kurzweil: 2029. Hinton: 20% chance before 2030.
Metaculus / Kurzweil / HintonWhat some prominent researchers have publicly said.
Paraphrased summaries below, not direct quotes. Views shown are not unanimous across the field - they are a sample of public statements that have been influential in the discourse.
Paraphrased from public statements, not direct quotes. Follow source links to read originals. Read our disclaimers and terms.
Has publicly argued that AI capability is accelerating quickly enough that AI-capable AI research could plausibly arrive within a few years, with significant downstream implications.
Has publicly raised concerns about the long-term safety of systems that may eventually exceed human cognitive capability and argued the topic deserves serious attention.
Has long argued for specific predictions about when computers may reach human-level intelligence and what he terms the singularity.
Has publicly suggested that broadly human-capable AI for intellectual work may arrive on a relatively short timeline, while emphasizing safety work.
Has publicly described the emerging period as one of potentially historic technological transformation, with both upside and risk.
Has publicly emphasized that very powerful AI systems are being built and that safety should be taken seriously.
Recommended Reading
Situational Awareness by Leopold Aschenbrenner has been influential in the discourse around AI timelines. A former OpenAI researcher's argument for why he thinks rapid AI progress is likely. Read it and form your own view - it is one author's argument, not a settled forecast.
~2 hour read. Worth every minute.
What this means for planning
The expert disagreement here is mostly about speed, not direction. Nearly everyone studying this professionally agrees that AI capability will continue to grow significantly. The relevant planning question isn't whether to account for AI disruption. It's what timelines your decisions should assume, and how much lead time you have to adapt the systems and institutions you're responsible for.
Illustrative benchmark figures
AI performance on various benchmarks has improved substantially over recent years. The numbers below are illustrative reference figures, not a settled comparison - benchmarks have known limitations and results vary by methodology.
The information here is a starting point for your own research, not a professional recommendation. Figures are rough indicators drawn from public reporting; verify against the original source. Read our terms & disclaimers.
So what might one do with this?
One possible response: rather than only competing with AI for cognitive work, consider using AI as a tool while also investing in tangible capabilities. This is the author's framing, not a recommendation.
The same technologies changing knowledge work also lower the cost of building, learning, and producing. Whether that matters for any specific person depends on their situation. Not advice.
About this project's book
Note: The paraphrased positions above are from independent third parties and do not constitute endorsements of this project or its book. This section is about the author's own work, separate from the views shown above.
Material Independence
What's inside:
- 01 Economic analysis - the author's reading of recent cost trends
- 02 The Shouse model - one approach to lowering housing overhead
- 03 Creator work as a possible income channel (outcomes vary)
- 04 Local AI tools and the case for running them yourself
- 05 Land and construction notes (consult your local attorney and contractor)
- 06 Open-source automation projects worth knowing about
- 07 Small-product business notes (not a guaranteed income model)
- 08 A suggested sequence the author has been thinking about
The core blueprint is free to read online. The book is an extended treatment of the same material - informational only, not advice. See full disclaimer.
People are already building.
Two paths. Same goal. Very different math.
- $500K house → $1.1M after interest
- 30-year commitment
- Requires steady employment
- Dependent on employer, market, economy
- No productive capacity
- $5-30K land + $25-100K shop/shouse
- 2-5 year build timeline
- YouTube funds the build in real-time
- Own your land, tools, and AI stack
- Productive from day one
The window is open.
Start building.
You don't need to quit your job tomorrow. You don't need to buy land this week. But you do need to start. Read the blueprint. Run the numbers. See if this path makes sense for you.
The blueprint is free. Always will be. The book is for people who want to go deeper.
Still skeptical? See the primary sources: footage and papers you can check yourself.