Paper and experiment

Geometry Before Language

A new way of making language models reason from computed structure — and a simple experiment designed to find out whether birth geometry contains any information about the person born beneath it.

“I would rather know than be right.”

The Chronometer calculates the positions and relationships of the planets using astronomy. It does this before a language model is allowed to say anything.

The model cannot move a planet, invent an aspect or alter the underlying geometry. It can only describe the structure it has been given. This approach is called Geometric Symbolic Reasoning.

The paper explains the architecture. The accompanying experiment asks a much simpler question: when people are shown eleven anonymous descriptions, can they recognise the one generated from their own birth information?

Two documents

Read the idea and the test

The paper

A proposed reasoning architecture in which structure is computed first and language is allowed to describe it afterwards.

The paper separates two claims. The first is that Geometric Symbolic Reasoning is a useful way to constrain artificial intelligence. The second is the open question of whether astrological geometry contains measurable information.

Explains the Chronometer architecture
Reviews major negative astrology studies
States the strongest sceptical explanation
Proposes a falsifiable test

Read Geometry Before Language

The experiment

Eleven descriptions are shown to each participant. Only one was generated from their actual birth date, time and location.

The other ten descriptions come from matched, astronomically possible birth moments. Their order is shuffled and all identifying clues are removed. The participant must decide which description is theirs.

One real interpretation
Ten matched control interpretations
No biography supplied to the system
A clear chance result of 1 in 11

Open the experiment design

The experiment in plain English

Step one

You enter the date, time and place where you were born.

Step two

The system generates your description and ten convincing alternatives.

Step three

You read all eleven and choose the one that sounds most specifically like you.

1 in 11

Pure guessing should identify the correct description about 9.09% of the time. If a large group consistently performs above that rate, there is something to investigate. If they do not, the hypothesis has failed this test.

What a positive result would not prove

Success would not prove that planets cause personality, that astrology is universally true or that a supernatural force is involved. It would only show that the correct birth geometry produced descriptions people could identify above chance. Replication and investigation would then begin.