Dev.to AI 🤖 Ai 👁 0

Four personas voted four to none, and it was still one opinion

When researchers gave ChatGPT 3.5 Turbo personas built from real election survey respondents, 48 percent of the regression coefficients differed significantly from the humans. The averages tracked people closely, but t

When researchers gave ChatGPT 3.5 Turbo personas built from real election survey respondents, 48 percent of the regression coefficients differed significantly from the humans.

  • The averages tracked people closely, but the synthetic answers varied less, and on 32 percent of the coefficients that differed, the sign flipped. Rerunning the same prompt after a model update did not reproduce the first result.
  • My manuscript cabinet voted four to none on a chapter, on one model, reviewing a draft the same run wrote. Every lens volunteered the same doubt about its own verdict, in the same direction.
  • So I asked a different model maker. My outside checker has five verdicts on record, five dissent and none concur, so I cannot show it would ever call anything sound.

Read the full article: What a Synthetic Persona Cannot Tell You

This summary was written with AI assistance (Claude) from the full article at https://www.orbytlabs.ai/blog/what-a-synthetic-persona-cannot-tell-you.

📰 Read the original article on Dev.to AI

Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.