---
date: '2025-12-23'
description: and strong longtermism.
id: singularity hypothesis
modified: 2026-06-05 15:08:22 GMT-04:00
tags:
  - pattern
  - philosophy
title: singularity hypothesis
created: '2025-12-23'
published: '2025-12-23'
pageLayout: default
slug: thoughts/singularity-hypothesis
permalink: https://aarnphm.xyz/thoughts/singularity-hypothesis.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
The singularity hypothesis says forecasts become unreliable after superhuman [[thoughts/AGI|AI]] appears. Vinge’s version is an event horizon: once humans create minds smarter than humans, we may lose the ability to predict the world those minds build. Good supplies the recursive mechanism. An ultraintelligent machine designs a better machine, and the next step starts from there.

"[[thoughts/papers/turing1950.pdf|Computing Machinery and Intelligence]]" matters for a different reason. Turing replaced “Can machines think?” with the imitation game, a behavioural test over written conversation. Passing that test would be evidence about human-like performance. It would not prove recursive self-improvement, takeoff speed, or longtermist stakes.

For LLMs, the question is concrete: can the system run the loop that improves it? The loop spends compute, collects data, runs experiments, edits code, and evaluates a successor. The hypothesis gets stronger as each step needs less human time.

![[thoughts/LLMs#^emergent]]

re [Vinge 1993](https://edoras.sdsu.edu/~vinge/misc/singularity.html), [Good 1965](https://incompleteideas.net/papers/Good65ultraintelligent.pdf), and [Turing 1950](https://www.hec.edu/sites/default/files/documents/Computing%20Machinery%20and%20Intelligence%20by%20Alan%20Turing.pdf)