The receipts
The Evidence
Everything on this site is downstream of things you can watch or read yourself. No forecasts you have to take on faith. Below is the footage and the primary sources: 8 demos and 9 papers and reports, organized by theme. Watch it, read it, pull it apart.
The timeline
What already happened, what is happening now, and the dates the people building it have put on the record.
- 2017
Attention Is All You Need
Google publishes the Transformer. The architecture behind every modern LLM.
Source: NeurIPS · arXiv:1706.03762 - 2022–24
The LLM era goes mainstream
ChatGPT, Claude, and Gemini put general-purpose AI in hundreds of millions of hands.
- 2024
"Situational Awareness" maps the decade
A short-timeline case for AGI by ~2027 starts shaping how labs and governments plan.
Source: situational-awareness.ai - 2025–26
Bodies, worlds, and new substrates
Humanoids hit production lines, models generate explorable worlds, neurons compute, and quantum chips stabilize.
Source: Google DeepMind - 2028
NASA SR-1 launches for Mars
The first nuclear-reactor-powered spacecraft is slated to depart by the end of 2028.
Source: MIT Technology Review - 2029
A useful quantum machine
Microsoft's stated target for a commercially scalable quantum computer.
Source: Microsoft Source - 2030
First commercial fusion plant
The IEA's milestone for fusion crossing from experiment to grid.
Source: Fusion Industry Association
2017 · where it began
The Spark
It did not start with a chatbot. It started with one paper that taught machines to pay attention.
Intelligence
Minds
Machine intelligence is going general: reasoning, planning, and now generating entire worlds to act inside.
Situational Awareness: The Decade Ahead
A former OpenAI researcher argues that AGI is plausible by 2027 and that the jump to superintelligence could follow fast, triggering a trillion-dollar compute buildout and a national-security scramble.
Why it matters: The most-cited articulation of the short-timeline view. Read it to understand why "before 2027" keeps coming up.
Open sourceFrom AGI to ASI
DeepMind researchers map four routes from human-level AI to superintelligence (scaling, paradigm shifts, recursive self-improvement, and multi-agent systems) and argue the transition is likely a series of shocks rather than one moment.
Why it matters: The lab actually building it, reasoning in public about what comes after AGI. Not a skeptic and not a hype account. The engineers.
Open sourceEmbodiment
Bodies
Intelligence is getting hands. Humanoid robots are already on factory floors and starting on the chores at home.
Access
The Workshop
The capability is not staying locked inside big labs. Industrial-grade tools are collapsing in size and price onto ordinary desktops, and capable robots now cost four figures, not six.
AhaRobot: a low-cost open-source bimanual mobile manipulator
A full dual-arm mobile manipulation robot you can build for about $1,000 (≈$2,000 with onboard GPU), roughly 1/15 the cost of a comparable $32,000 commercial platform.
Why it matters: The $100k robot arm is a myth now. The plans and the bill of materials are public, and the total is four figures.
Open sourceMobile ALOHA: learning bimanual mobile manipulation
Put the low-cost two-armed rig on a mobile base and teach it by demonstration: it learns to cook, store items in cabinets, ride elevators, and clean a kitchen.
Why it matters: A trainable, low-cost robot chef is not science fiction. It is a Stanford demo from 2024 with the recipe published.
Open sourceALOHA 2: enhanced low-cost hardware for bimanual teleoperation
An open-source, low-cost two-armed platform built for large-scale manipulation data collection, with improved grippers, frame, and cameras over the original ALOHA.
Why it matters: Top labs are deliberately driving the cost of capable robot hardware down and open-sourcing it. The barrier to entry is falling on purpose.
Open sourceHardware
New Substrates
The stuff underneath is changing too: living neurons doing computation, and a real path toward stable quantum chips.
Energy
Power
None of it runs without energy. Solar already collapsed in price; fusion is now on the near-term map.
Frontier
Off-World
The same curve does not stop at the atmosphere. Nuclear propulsion puts Mars inside a single-year trip.
What it adds up to
We are already in the early stages
Pull back from any single demo and the pattern is hard to miss. Intelligence, bodies, cheap tools, new substrates, and abundant energy are all bending upward at once, and AI is now helping build the very hardware that runs the next AI. That feedback loop has a name: recursive self-improvement. The point where machines improve themselves faster than we can, at scale, indefinitely.
The question is no longer if recursive self-improvement happens. It is when, and how fast.
Soft takeoff
The transition plays out over years. Capability compounds gradually, society and institutions have time to adapt, and the change feels like a steep but climbable ramp.
Hard takeoff
The transition happens in months or weeks once the loop closes. Capability goes near-vertical, adaptation time collapses, and most people only notice after the world has already changed.
The evidence above keeps pointing the same direction: a hard takeoff looks increasingly likely. That is not a reason to panic. It is a reason to build your independence before the curve goes vertical, not after.
How to use this page
- Skeptical? Good. Every claim links to its source. Check them.
- Want to share it? Send the link. It is one page, no signup, no paywall.
- Talking to an AI? Hit "Copy page for chat" up top, paste it in, and ask it whatever you want.
- Think something is missing or wrong? That is the point. The list is meant to be argued with.