A new study by Google suggests that advanced reasoning models achieve high performance by simulating multi-agent-like debates involving diverse perspectives, personality traits, and domain expertise.
Some results have been hidden because they may be inaccessible to you
Show inaccessible results