Fascinating. The Deep Think explanation was truely eye-opening. It's brilliant how they're tackling reasoning in paralel. Such a smart approach to boost performance!
When the price of tokens goes down enough, then more interesting stuff, like a massive tree of thoughts, become feasible. The model can then explore in depth multiple ideas, verify each step and find more optimal or unexpected solutions to a problem. DeepMind and others might have already looked into it, but it might be too expensive to serve it at scale.
Fascinating. The Deep Think explanation was truely eye-opening. It's brilliant how they're tackling reasoning in paralel. Such a smart approach to boost performance!
When the price of tokens goes down enough, then more interesting stuff, like a massive tree of thoughts, become feasible. The model can then explore in depth multiple ideas, verify each step and find more optimal or unexpected solutions to a problem. DeepMind and others might have already looked into it, but it might be too expensive to serve it at scale.