I want to write a few words to the community about an issue you may have encountered with our Muse Chat where Muse’s response gets cut off or truncated during conversations. The issue that I keep hearing is that some folks will be deep in a conversation and once the truncation starts, it continues to happen until starting a new Chat, which is frustrating because you lose the prompt/response context built when you start a new Chat.
The reason this is happening is due to the model’s token capacity. There is a cap on how much historical data the model can retain at once. Responses may be cut off mid-stream if the available token space for generating responses hits a limit. Right now we’ve created a solution where we essentially reserve token space for the response Muse generates, and as you continue conversations, historical context is dynamically managed to accommodate new messages. Consequently, longer conversation histories may increase the likelihood of mid-stream truncations.
We understand this is frustrating, and the team has been working hard on trying to find a solution that mitigates this and optimizes this process for smoother interactions. We appreciate your patience and understanding as we work through this! The team has been extremely busy this quarter - they also just rolled out some changes to the UI to address the user pains of shortage of vertical height, too dark, and adding more discoverability to the side panels. They’re super! So I have no doubt we’ll be able to collaborate to mitigate this.
Thank you for your continued support as we try to make the best Unity assistant for unblocking obstacles during game development. If you have any further questions or concerns, please don’t hesitate to contact our user support.
Best,
Liz