More Agents Is All You Need: LLMs performance scales with the number of agents
More Agents Is All You Need: LLMs performance scales with the number of agents
arxiv.org /abs/2402.05120
2
comments
Good, people kept complaining that they don't use enough power. Now we can have more agents working simultaneously on the same task so we can use 3x, 5x, or maybe 10x the power usage! 🎉
I think this was already happening though, to be fair. The outlined technique actually cuts away some amount of work for the more commonly used multiple agent technique.