Flutterby™! : careful what you tell those models

Honestly, this makes sense: Programmers who proclaim themselves expert are generally crappy programmers, so if you tell an LLM that it's supposed to be an expert programmer, of course it's gonna generate worse code. The Register: Telling an AI model that it’s an expert programmer makes it a worse programmer

In a pre-print paper titled "Expert Personas Improve LLM Alignment but Damage Accuracy: Bootstrapping Intent-Based Persona Routing with PRISM," researchers affiliated with the University of Southern California (USC) find that persona-based prompting is task-dependent – which they say explains the mixed results.

Via.