It's not bad for something that's pretty new. Now imagine what another decade or two of development will give us. The number of areas in which a typical worker is more knowledgeable is going to get lower and lower.
It can already ingest 80kb of information instantly and use it to perform any knowledge task that can be described in that amount of info (with the 32k token model).
The number of areas in which the typical worker is better than a GPT3.5 or GPT4 system that has been properly integrated with the correct data and tools is already probably less than 5% for knowledge workers.