Google study shows LLMs abandon correct answers under pressure, threatening multi-turn AI systems

A DeepMind study finds LLMs are both stubborn and easily swayed. This confidence paradox has key implications for building AI applications.

Raccontata daventurebeat.com