> ... the target size for the new model is 1 trillion parameters.
> Meanwhile, the target size for the free & public versions of ChatGPT is just 175 million in comparison. That's a 5.7x increase in number of the parameters.
5.7x seems off by several orders of magnitude, or do I need more caffeine?
> Meanwhile, the target size for the free & public versions of ChatGPT is just 175 million in comparison. That's a 5.7x increase in number of the parameters.
5.7x seems off by several orders of magnitude, or do I need more caffeine?