---
schema: 1
type: term
title: '**Big Five traits**'
description: >-
  **Big Five traits** — in this channel's material, a set of personality
  dimensions used as the benchmark for how accurately a language model can
  simulate a real person: a persona built from a two-hour interview transcript
  matched the actual person at roughly **80% accuracy on Big Five traits,
  economic-game decisions, and voting** — about as consistent as the person is
  with their own past self. [^01a0f15a-e590-7f2d-8b7d-8afcf1ca33f3@408]
canonical: >-
  https://liquidgold.cc/channel/everyinc/explore/term/big-five-traits-acd07d3118a4
markdown: >-
  https://liquidgold.cc/channel/everyinc/explore/term/big-five-traits-acd07d3118a4.md
llms: null
parent: 'https://liquidgold.cc/channel/everyinc/llms.txt'
channel: everyinc
updated: '2026-10-23T22:12:35.346Z'
language: null
children: []
---


# \*\*Big Five traits\*\*

Canonical: https://liquidgold.cc/channel/everyinc/explore/term/big-five-traits-acd07d3118a4

Parent: [Channel index](https://liquidgold.cc/channel/everyinc/llms.txt)

## Definition

**Big Five traits** — in this channel's material, a set of personality dimensions used as the benchmark for how accurately a language model can simulate a real person: a persona built from a two-hour interview transcript matched the actual person at roughly **80% accuracy on Big Five traits, economic-game decisions, and voting** — about as consistent as the person is with their own past self. [^01a0f15a-e590-7f2d-8b7d-8afcf1ca33f3@408]

In this channel's material, **Big Five traits** appear in the context of simulating audiences with language models. Michael Taylor (Rally) cites research showing that if you take a **two-hour interview transcript**, build a personality profile from it, and query the resulting simulated person, that simulation matches the real person at roughly **80% accuracy on Big Five traits, economic-game decisions, and voting**. [^01a0f15a-e590-7f2d-8b7d-8afcf1ca33f3@408]The kicker Taylor emphasizes: that ~80% figure is about as accurate as *interviewing the same person a second time* — because humans themselves only agree with their past selves about 80% of the time. As Dan put it during the conversation:> “I contain multitudes.”This makes Big Five traits a **benchmark for persona fidelity**: they are one of the measurable dimensions along which a language-model-generated persona can be checked against the real person it simulates. The practical takeaway for the channel's broader prompt engineering theme — roleplay and audience simulation work well enough to ground real decisions, but with honest limits, since prediction diverges wildly beyond the near horizon. [^01a0f15a-e590-7f2d-8b7d-8afcf1ca33f3@244]- **Big Five traits** show up as the accuracy benchmark for personality-based persona simulation: ~80% match from a two-hour interview transcript. [^01a0f15a-e590-7f2d-8b7d-8afcf1ca33f3@408]- That accuracy is roughly what you get from re-interviewing the same human — humans only agree with their past selves ~80% of the time.- The context is Michael Taylor's argument that simulating audiences with language models turns market research into a cheap, repeatable experiment.