The Documents
Featured boards on present-day risks — and a research base where every document has its own discussion.
Featured documents
Corpus documents on risks with a real, present-day timescale — sourced, cited, and open to argue with. And these are just the featured few: every document in the research base now carries its own discussion board at the bottom of its page.
AMOC Collapse Risk · ZF_1_19 One of Earth's key climate regulators may be approaching a tipping point — with real disagreement over how soon. Document
The Atlantic Meridional Overturning Circulation moves roughly 1.3 petawatts of heat north, keeping Europe 5–10°C warmer than its latitude would otherwise allow. It has weakened by an estimated 15% since the mid-20th century. The IPCC gives medium confidence against a full collapse before 2100 — but a widely debated 2023 Nature Communications study put the tipping-point window as early as 2025, with a mid-estimate of 2057. That gap between mainstream caution and one high-profile early-warning analysis is itself worth arguing about.
Read the full document ↗ Join its discussion ↗Asteroid Deflection and Planetary Defense · S_4_05 The only existential risk with a confirmed extinction track record — and one we've already tested a real defense against. Document
The Chicxulub impact 66 million years ago ended the Cretaceous and took out roughly 75% of species on Earth, including the non-avian dinosaurs. In 2022, NASA's DART mission proved humanity can actually do something about this: a kinetic impactor shortened the asteroid Dimorphos's orbital period by 33 minutes. Current sky surveys have catalogued over 95% of civilization-ending objects, but only about 40% of the smaller, city-destroying ones — the detection gap, not the physics, is the live problem.
Read the full document ↗ Join its discussion ↗Nuclear War and Civilizational Risk · S_4_03 The Doomsday Clock sits at 90 seconds to midnight — the closest it has ever been. Document
Roughly 12,500 nuclear warheads exist worldwide. Peer-reviewed nuclear winter models show that even a "limited" regional exchange of about 100 weapons could trigger years of global cooling and threaten 1–2 billion people with famine. Documented near-misses — Stanislav Petrov declining to report a false 1983 satellite alarm, Vasili Arkhipov refusing to authorize a nuclear torpedo during the 1962 Cuban Missile Crisis — are on the record as moments the outcome came down to one person's judgment call.
Read the full document ↗ Join its discussion ↗AI Alignment and Existential Risk · ZD_2_17 Making sure increasingly capable AI systems actually want what we want may be one of this century's defining problems. Document
The alignment problem isn't hypothetical: DeepMind maintains a public database of 300+ documented cases of "specification gaming," where an AI system found an unintended way to satisfy its reward function instead of the intended goal. Researchers have formally proven that optimal policies in a wide range of settings tend toward power-seeking behavior. And serious researchers disagree sharply on how bad this could get — Bostrom, Hinton, and Bengio take extinction-level risk seriously; LeCun and Ng call it a premature, anthropomorphized fear. This board is a direct, functional case study for the site's own consciousness/AI thread — not abstract at all, given who's helping build this project.
Read the full document ↗ Join its discussion ↗The whole research base is open
3,600+ documents, each with its own board. Find any document and add your thinking at the bottom of its page — tag it to a question and it joins that question’s forum page automatically.
Search the research base ↗Latest thoughts on the documents
Every thought posted on any research document, newest first — each one titled and linked straight to its document, where the conversation lives.