The breadth and formatting of information has changed so that models are able to store more relevant information about the user and respond appropriately, this includes recent conversations.
Memory is now searchable by the AI assistant
Previous Conversations are now searchable and accessible to the model (this can be disabled individually)
When needed, the model can view all memories stored for the user, to provide a comprehensive view when needed, but avoids doing so unless necessary.
You can now cancel messages mid-generation if needed, you will still be charged credits if the model has started generating content
Messages that fail now display an error message
Bugs Fixed:
Fixed a bug that caused conversations containing both multimodal and non-multimodal models to fail under specific circumstances