@Alex Kantrowitz is it really a “new model” tho? or makes any advancements in model architecture? from what it seems, it is an existing model with a recursive prompt structure. It’s kinda a multi-step agenetic approach to CoT that saves some time iterating with the model. This is not to say it’s not a step ahead, but I don’t think it’s an advancement in the model space itself.
It is not a new model. It is GPT-4o with Chain of Thought reasoning built into it via reinforcement learning. In fact, internally it seems to be referred to as “GPT-4o with Reasoning”. That said, I do think it is a step forward and shows that model scaling isn’t the only way to advance performance
@Alex Kantrowitz is it really a “new model” tho? or makes any advancements in model architecture? from what it seems, it is an existing model with a recursive prompt structure. It’s kinda a multi-step agenetic approach to CoT that saves some time iterating with the model. This is not to say it’s not a step ahead, but I don’t think it’s an advancement in the model space itself.
It is not a new model. It is GPT-4o with Chain of Thought reasoning built into it via reinforcement learning. In fact, internally it seems to be referred to as “GPT-4o with Reasoning”. That said, I do think it is a step forward and shows that model scaling isn’t the only way to advance performance