Update OFRAK AI with the increased history, improved structured output from recent API improvements

[marczalik]

OpenAI recently rolled out changes to their API that drastically increases the history for the 3.5 model, which should improve results for long-running conversations. In particular, this should help maintain contextual information when sending lots of data to the model, such as in the program/function analyzers.

These API changes should also improve the "steerability" of the model's output, creating more reliably structured output that we can use. This could help, for instance, in making sure string format specifiers and response lengths are respected.

News article here. Announcement here.