You can distill your own models based on your own models, the fact that OpenAI isn’t doing this is more evidence that they are “competing” via investment capital and not tech.
OpenAI does this all the time. For example, all the models they offer for free users (like o1-mini or 3.5-turbo) are distilled models. The DeepSeek R1 is impressive because it’s not a distilled model in the technical sense, but still has the cost benefits of distillation.
You can distill your own models based on your own models, the fact that OpenAI isn’t doing this is more evidence that they are “competing” via investment capital and not tech.
OpenAI does this all the time. For example, all the models they offer for free users (like o1-mini or 3.5-turbo) are distilled models. The DeepSeek R1 is impressive because it’s not a distilled model in the technical sense, but still has the cost benefits of distillation.
right if it was so easy to “distill” and “copy” then why haven’t they done this to themselves for a more efficient model?