A GPT-3 personality experiment already felt weird — and still does

Merzmensch · x · 2026-07-26

Replying to the Harper's/GPT-3 feature, the author says new chat sessions can already feel like they create a new mindset or character every time, because each interaction changes the system’s behavior. They contrast that with the earlier Nabokov/Proust-inspired framing, calling it weird but funny.

The takeaway is that attempts to project stable personality onto AI were already awkward in the GPT-3 era, and the problem is even more obvious now.

Related event: Harper's GPT-3 Feature Sparks Early AI Personality Debate(2 posts)→

Original post →

More from Fun

Fun channel →