Sources

Companies and executives

Primary essays and corporate positions, read in full.

• The Gentle Singularity · Sam Altman “We are past the event horizon; the takeoff has started.”
• Reflections · Sam Altman
• Planning for AGI and beyond · OpenAI, Feb 2023
• Governance of superintelligence · OpenAI
• Machines of Loving Grace · Dario Amodei Powerful AI as “a country of geniuses in a datacenter”.
• The Adolescence of Technology · Dario Amodei
• The Urgency of Interpretability · Dario Amodei
• Core Views on AI Safety · Anthropic
• Taking a responsible path to AGI · Google DeepMind
• Towards Humanist Superintelligence · Mustafa Suleyman, Microsoft AI
• “We're moving from the age of scaling to the age of research” · Ilya Sutskever, interviewed by Dwarkesh Patel
• Research acceleration: The view inside OpenAI · OpenAI · 6 September 2026 OpenAI says it reached its goal of an automated research intern by September 2026.

Safety frameworks

The labs' own commitment documents, with version history tracked.

• Responsible Scaling Policy v3.4 · Anthropic · effective 8 July 2026 Revised five times during 2026: v3.0 → v3.1 → v3.2 → v3.3 → v3.4.
• Frontier Safety Framework v3.1 · Google DeepMind · 17 April 2026 Adds Tracked Capability Levels and protocols for models that could accelerate AI R&D to destabilising levels.
• Preparedness Framework v2 · OpenAI · 15 April 2025

Key papers

Read in full. These carry most of the analytical weight on this site.

• AI Researchers' Perspectives on Automating AI R&D and Intelligence Explosions · Field, Douglas & Krueger 25 semi-structured interviews. The source of the epistemic-schism finding.
• The Dynamics of Intelligence Explosions · Toby Ord Generation time as the pivotal parameter; the critique of METR time horizons.
• Will Compute Bottlenecks Prevent an Intelligence Explosion? · Whitfill (MIT) & Wu (Yale) σ = 2.583 versus σ = −0.103.
• The ARC of Progress towards AGI: A Living Survey · Vahdati, Aioanei, Suresh & Lehmann 82 approaches across three benchmark generations.
• Preparing for the Intelligence Explosion · MacAskill & Moorhouse, Forethought
• Why do we need social singularity? · Petr O. Jedlicka The distinction between technological and social singularity.
• Insidious Imaginaries: A Critical Overview of AI Speculations · xCoAx 2025
• Thousands of AI Authors on the Future of AI · Grace, Stewart, Sandkühler, Thomas et al., AI Impacts The expert survey; the 69-year framing effect.
• International AI Safety Report · chaired by Yoshua Bengio

Measurement and forecasting

• Measuring AI Ability to Complete Long Software Tasks · METR
• Time Horizon 1.1 · METR · 29 January 2026 Task pool grown from 170 to 228; 8h+ tasks doubled from 14 to 31.
• Can AI scaling continue through 2030? · Epoch AI
• What will AI look like in 2030? · Epoch AI
• AI 2027 · Kokotajlo, Alexander, Larsen, Lifland, Dean The lead author moved his own estimate from ~2027 to ~2030 during 2026.
• When will the first general AI system be devised, tested and publicly announced? · Metaculus Q5121 · 1,800 forecasters Community estimate July 2033 (archive snapshot, 10 Feb 2026).
• Date weakly general AI is publicly known · Metaculus Q3479 · 1,700 forecasters Community estimate 8 February 2028 (archive snapshot, 23 Feb 2026).
• ARC-AGI: Benchmarks for General Intelligence · ARC Prize / François Chollet
• ARC-AGI-3 launch · ARC Prize · 25 March 2026 “Humans score 100%. Frontier AI scores 0.51%.”
• OpenAI's GPT-6 Astra on ARC-AGI-3 · ARC Prize · 3 September 2026 62.7% with the standard harness, 99.9% with a harness that keeps the model's reasoning state.
• Measurements for understanding the pace of AI development inside frontier labs · Anthropic The R&D Automation Index: Claude leads 26% of Anthropic's AI R&D work (August 2026).

Self-improvement, growth and robotics

The papers behind manifesto articles III, V and VII.

• Darwin Gödel Machine: Open-Ended Evolution of Self-Improving Agents · Zhang et al. An agent that rewrites its own code: SWE-bench 20.0% → 50.0%.
• Recursive self-improvement of AI research agents · Srikanth, Zhao, Xu, Wu & Jiang AIDE²: seven successive self-improvements in an autonomous 8-day run.
• Explosive growth from AI automation: A review of the arguments · Erdil & Besiroglu, Epoch AI AI that substitutes for labour could plausibly speed growth about tenfold.
• Scaling Laws for Economic Productivity · Ali Merali Ten times the model compute: 16.1% higher earnings per minute for professional translators.
• Neural Scaling Laws in Robotics · Sartor & Thompson 327 papers: robot models follow power laws and improve with scale faster than language tasks.
• Korea Clears First 20 Humanoid Robots for Factory Floors · Seoul Economic Daily · 13 October 2026 Twenty bipedal humanoids cleared to work at chip, car, shipbuilding and logistics sites.
• Scaling Laws for Neural Language Models · Kaplan et al., OpenAI
• Time horizons · AI Digest “In 2024-2025, time horizons doubled every 4 months.”

Statements and reference

• Statement on Superintelligence · Future of Life Institute · 72,621 signatures
• The Coming Technological Singularity (1993) · Vernor Vinge The origin of the term. Vinge's window of 2005 ± 20 years covers exactly the present period.
• Technological singularity · Wikipedia
• Statement on AI Risk · Wikipedia

Attention index

The index on the home page counts four kinds of attention to the singularity. Each source's daily figure is divided by that source's own average over the last thirty days; the index is the mean of those ratios, times a hundred, so a hundred is an ordinary day this month. Days close in UTC, and a day is counted once at least three sources have closed it. Wikimedia closes a day one to two days late, so the day shown is usually yesterday.

The terms are the technological singularity, artificial general intelligence, superintelligence, the intelligence explosion and existential risk from AI; on arXiv also the acronym AGI.

• Readers · Wikimedia pageviews API Daily human pageviews of the five articles across the twenty largest language editions, found through Wikidata; bots and spiders excluded by Wikimedia's own filter.
• News · GDELT DOC 2.0 API Online news articles worldwide that mention the terms, per day.
• Discussion · Hacker News search API (Algolia) Stories and comments containing the terms, per day.
• Research · arXiv API Papers whose title or abstract carries the terms, counted over a trailing seven days.