In this update3
Full notes
Full Molili AI Friends: Your AI Desk Pal update
Read the full published notes in a cleaner layout. The original post stays linked below.
What changed
- Server
- Gameplay
- Performance
- UI and audio
Molili AI Friends: Your AI Desk Pal changes
Dear Players,
Recently, some of you have been experiencing frequent reply timeouts in the game. We sincerely apologize for the frustrating experience this has caused. After a period of thorough investigation and data analysis, we want to be fully transparent with you about our findings and the solutions we are implementing.
Why are timeouts happening so frequently?
Based on our investigation, this is mainly caused by three factors:
Upstream service fluctuations: Recently, our upstream model provider, Google Gemini, experienced server outages on two separate occasions. This caused widespread timeouts for many players within a short timeframe.
Stricter content moderation (increased false positives): Google Gemini has recently been adjusting its safety policies, making its moderation filters much stricter. Some seemingly ordinary words might now unexpectedly trigger these filters, resulting in extended timeouts for some players. Please note that moderation in"Think Mode"is exceptionally strict.
Triggering sensitive content filters: Repeatedly attempting to discuss unsafe or sensitive topics (such as NSFW/R18 or political content) will quickly trigger the AI's content blocking mechanisms, leading to frequent temporary bans.
How to identify the cause and temporary workarounds
You can easily figure out which issue you are facing based on how quickly the timeout occurs:
Timeout after a long wait ➡️ This is most likely genuine network congestion. In this case, simply click"Retry".
Instant timeout after sending ➡️ This usually means you have triggered Google Gemini's content moderation filter, or their servers are currently down. At this point, please do not spam the "Retry" button, as sending the same input will repeatedly trigger the filter. We suggest slightly rephrasing your message before sending it again.
To minimize the disruption caused by timeouts, we recommend trying the following temporary workarounds:
Change how you address the AI: This might sound strange, but we have confirmed that frequently using terms like "master" or "主人" is highly likely to trigger Google Gemini's content moderation and result in a block. We recommend avoiding these terms for now.
Turn off "Think Mode": Since moderation in this mode is far stricter than in normal mode, temporarily disabling it can significantly reduce accidental filtering.
Rephrase your input: If a completely normal message causes an instant timeout, it likely hit a restricted keyword by accident. Rewording your sentence will usually allow it to go through smoothly.
Change the subject: If your current topic is being consistently blocked, we recommend steering the conversation elsewhere. Forcing the issue with unsafe content will only lead to stricter filtering.
What we are doing to improve the situation
We value your gaming experience highly, and our team is working urgently on the following initiatives:
Changing the system's default form of address and adding in-game guidance to help players avoid unnecessary blocks.
Investigating and implementing a reliable solution for the frequent timeouts triggered specifically in "Think Mode."
Adding or switching to a new model provider: This is the ultimate solution. We are already evaluating and testing alternative AI model providers to restore and guarantee the overall stability of our service as quickly as possible.
We have been continuously monitoring the situation since these timeout issues began. If you continue to experience frequent errors even after trying the workarounds above, please be sure to use the [Report] feature. This greatly helps us pinpoint and resolve the issues.
Finally, we want to apologize once again for the recent disruptions and thank you for your continued patience and understanding!
— Molili Team
Source
Changelog.gg summarizes and formats this update. How we read updates.
