there’s now an open-source project called AutoGPT that enables OpenAI GPTs 3.5 and 4 to break apart tasks, remember, and interface with the internet, which allows them to get current data and perform multi-stage tasks. it still has a lot of problems with focus and complex goals, but that will only get better as time goes on.
additionally, researchers created a small town of autonomous AI agents who were provided with backstories and were able to interact with each other and their environment in a game engine reminiscent of the RPG genre. this shows off the capabilities of even current AI models and could be used to make less important game NPCs much more interesting, unique, and real.
https://arstechnica.com/information-tec ... -rpg-town/