It’s the same in terms of conversation. Primarily just enhances its capabilities. Here’s an experiment I documented without using the continuous memory concept. The memory gives it more context, and also enables it to evaluate its progress toward a goal.
https://fullstackpete.com/this-is-incredible-gpt-4-untapped-potential/
Turning it into a full fledged AI entity is more of a challenge, but it’s definitely an interesting journey. It’s actually able to code, iteratively building on each snippet it adds, and is able to write its own prompt for the next ‘action cycle’. It’s wild stuff.