On: More Agents Is All You Need: LLMs performance scales with the number of agents

[dmarchand90]

Be interesting to plug this into a bayesian optimization like framework: find out regions of language space where the models maximally disagree and then target those areas for extra training