Singularity Countdown

version-1.3

An evidence-informed singularity clock using Bayes' theorem.

Status Log

  • System log:
  • Time to singularity: unknown
  • Fault cause source: black box
  • RSI status: AI-assisted self improvement observed
  • AC system online: potential early phase
  • Recursive improvement loop state: monitoring
  • Epistemology: Unknown
  • Model velocity mode: exponential acceleration
  • Temporal compression observed: 48h → 6h
  • Projection intervals: 1y → 6m → 3m → 1m → 1w
  • Rev 13:15 countdown: unknown
  • Human in the loop status: declining
  • Economy: unknown (but mechanism design economics hinted)

// Single Inference Clock (SIC)

Plots how a single AI scenario’s initial belief (prior) and false-positive base rate shift into an updated posterior probability after weighing new evidence, converting that confidence into a countdown.

Feel free to screenshot and share this inference clock.

Active inference
Prior probability P(H) 28%
Evidence if true P(E|H) 78%
Evidence if false P(E|not H) 34%
47% Updated posterior
Inference clock
Generated from current posterior

Why Bayesian rather than absolute? Absolute claims hide uncertainty and make new evidence difficult to incorporate. Bayesian inference keeps confidence proportional to the evidence, respects base rates, and makes each update visible without pretending that a probability is proof.

The word “forecast” is intentional. Like a weather forecast, this is an informed estimate, not a certainty. Weather is shaped by countless changing conditions, so even the best forecast can be wrong.

// Inference Breakdown
Hypothesis
Prior
Evidence
Posterior
P(H|E) = P(E|H)P(H) / [P(E|H)P(H) + P(E|not H)P(not H)]. Inputs are scenario assumptions informed by trend proxies, not measured probabilities of singularity.

Since there is no observed singularity to measure directly (it has never occurred), the model uses observable AI trends to update beliefs about scenarios that might lead towards one. That doesn’t mean the model is arbitrary; the aim is to take an empirically informed approach while being open about what is observed, what is assumed, and what remains unknown. For example, a food delivery app estimates arrival from restaurant preparation, courier location and speed, route, and traffic. When the order is ready and the courier is nearby on a clear route, delivery soon becomes more likely. Each update shifts the likely wait without guaranteeing an exact arrival time.

// Cumulative Inference Clock (CIC)

Feel free to screenshot and share this inference clock.

Cumulative inference clock
Awaiting cumulative inference.
How cumulative inference works: Selected inference signals are combined to show their cumulative effect on the estimate, like a detective building a case from multiple pieces of evidence.
// Spotlight News

Ruby on Rails creator DHH gives up writing code by hand

After 25 years as a professional programmer, David Heinemeier Hansson says he has moved from hand-writing code to directing AI agents. According to THE DECODER’s report on his Rails World 2026 keynote, the Basecamp co-founder said he had not written code by hand since around March 2026—a reversal of his earlier scepticism about AI coding.

His argument is economic: he believes manual coding is becoming difficult to justify for most programmers and companies. Instead, he sees developers focusing on deciding what to build and guiding AI through the work. That is his assessment of the shift, not proof that manual coding has ended across the profession.

Read the report · Watch the keynote

Source: THE DECODER’s report and DHH’s Rails World keynote. Summary checked .

// Fun

Current rating, 2026 K ≈ 0.73 about 20 TW
Type I target K = 1.0 10,000 TW, 500x today

Humanity's path to a Type I civilisation

The Kardashev scale ranks civilisations by the energy they control. Earth scores about 0.73, meaning we use only a small fraction of our own planet's energy. Type I masters a whole planet, Type II a star, Type III a galaxy, Type IV the universe. Type V many universes, and Type VI reality beyond space and time. Those later ones come from pop science, not the original scale.

All lines are scenarios: Singularity (rising rate), RSI 26%/yr, AI (LLMs) 8%/yr, Old trajectory 2%/yr

Feel free to screenshot and share this Kardashev chart.

Illustrative paths to a Type I civilisation Four dotted scenarios start at a Kardashev rating of approximately 0.73 in 2026. Singularity reaches 1.0 first, followed by recursive self-improvement and AI. The old trajectory remains near 0.82 by 2125. These are speculative illustrations, not predictions; year spacing is schematic. 1.0 0.9 0.8 0.7 2026 2030 2050 2075 2100 2125 Singularity RSI AI (LLMs) Old trajectory, based on obsolete tech
Informed by informal Bayesian reasoning about observed trends. The Kardashev scale provides the scientific framework; the trajectories are speculative. WIP: better evidence and more rigorous inference coming soon. Less fun (boring!).

// How It Works

The equation might seem a bit complicated, but it just means updating how plausible a scenario is when new evidence arrives. Evidence that is more likely if the scenario is true than if it is false raises its probability; evidence that favours the opposite lowers it. The update also accounts for how plausible the scenario was to begin with.

Each scenario starts with a prior: how plausible the outcome seemed before the latest evidence. That evidence is then weighed by how likely it would be if the scenario were true, compared with how likely it would be if it were false. Bayes’ theorem combines these into a posterior probability.

P(H|E) = P(E|H) · P(H) / P(E) where P(E) = P(E|H) · P(H) + P(E|¬H) · P(¬H)
P(H)
Prior: initial belief that the scenario is true
P(E|H)
Probability of seeing the evidence if the scenario is true
P(E|¬H)
Probability of seeing the evidence if the scenario is false
P(H|E)
Posterior: updated belief after weighing the evidence

The posterior drives the inference clock: higher confidence produces a shorter countdown, ranging from 10 years at 0% to 6 months at 100%. The inputs are scenario assumptions informed by trends such as METR’s task-horizon measurements, not measured probabilities, so the clock shows how belief should change with evidence rather than predicting a specific date.

// Intuitive Examples
  • If a car won’t start, a dead battery is one possibility among several, alongside a faulty starter or a loose connection.
    • Rapid clicking makes a dead battery more likely because that sound often occurs when the battery cannot provide enough power.
    • A faulty starter can make a similar sound, so the clue raises the likelihood of a dead battery without proving the cause.
    • Checking the battery and connections gives you more evidence and may change which explanation seems most likely.
  • A phone that won't turn on may have a drained battery, but a damaged screen or charging fault could also be responsible.
    • A charging symbol after plugging it in makes a drained battery more likely.
    • No response may point to a damaged screen, cable, or charging port instead.
    • Each clue changes which explanation seems most likely, but does not prove the cause.

Hypothetical food delivery

You're waiting for a food delivery. Will it arrive within 15 minutes? Start with a guess, then update it when the restaurant marks it ready and the courier is nearby on a clear route. The percentages below are illustrative assumptions, not measured delivery statistics.

  1. 01Prior

    You're waiting for a food delivery. Our hypothesis H is that it will arrive within 15 minutes. Before the restaurant or courier updates, imagine 3 out of 10 similar orders arrive that quickly. That gives a starting belief, or prior, of 30%. This is an illustrative assumption, not a measured delivery rate.

  2. 02Evidence

    The restaurant marks the order ready, and GPS places the courier nearby on a clear route. This is strong evidence for prompt delivery. The 98% and 2% likelihoods are hypothetical, just to show the gist of Bayesian updating; real likelihoods should be calibrated from delivery history.

  3. 03Posterior

    Because this evidence strongly favors delivery within 15 minutes, Bayes' rule raises the illustrative chance from a 30% prior to a 95.5% posterior. The result is only as reliable as the assumed likelihoods, and is not certainty.

  4. 04Delivery ETA

    The app estimates delivery from restaurant preparation and the courier's GPS, speed, route, and traffic. Reliable GPS showing the courier nearby can raise the posterior for prompt delivery; weak signal or new delays add uncertainty, so it is never guaranteed.

// FAQ

What is the Singularity Countdown?

An interactive Bayesian forecast that turns evidence about AI progress into a countdown. The clock ranges from 10 years at 0% posterior confidence to 6 months at 100%.

What is the technological singularity?

A hypothetical point at which AI improves itself so quickly, for example through recursive self-improvement, that technological change becomes very hard to predict.

When will the singularity happen?

No one knows. The countdown is a probability-weighted projection that shows how belief should change as evidence arrives, not a prediction of a specific date.

How is the countdown calculated?

Each scenario starts with a prior probability. Bayes' theorem weighs how likely the evidence would be if the scenario were true versus false, producing a posterior that sets the clock.