We are well aware of your expectations regarding AI response quality, and we are continuously improving our quality control system based on reports of unsatisfactory responses to provide a better conversation experience.
Here are the details on the most frequently asked items regarding message quality.
1️⃣ The response length is shorter than expected
The length of an AI's response may vary depending on the conversation situation, model characteristics, and context (e.g., if previous conversations were short), and it does not guarantee a specific character count every time.
Even when using the same mode and settings, the response length may vary according to the AI's judgment, which is a basic characteristic of AI language models. Therefore, instances may occur where the response is slightly shorter than expected (e.g., 350 characters when 500 were expected).
However, if the response is completely empty or extremely short (1-2 sentences), please let us know through an unsatisfactory response report, and we will prioritize it for improvement.
2️⃣ Foreign languages are mixed in the response
Because AI language models learn multiple languages simultaneously, words or sentences in languages other than the set language may occasionally be included. This is especially likely if the character setting has foreign language elements or if another language is mentioned during the conversation.
It is technically very difficult to control the AI to respond 100% in a single language. We apologize for the inconvenience and are working on improvements to reduce foreign language mixing. Instances where character names or proper nouns are written in the original language, or only some words appear in another language, are considered normal responses.
However, if most of the response is output in a language other than the set one (e.g., the entire response is in English) or if the proportion of foreign language is high enough to make the content difficult to understand, please let us know through an unsatisfactory response report, and we will review it as a priority.
3️⃣ Similar responses are repeated
The AI generates responses based on the conversation context, character settings, and recent conversation history. If you send messages with similar content in similar situations, the AI is likely to generate similar responses. Therefore, instances where expressions or nuances are similar in similar situations, or where a character's tone or habitual expressions are repeated (character personality), are considered normal responses.
If you want to avoid repetitive responses, try this:
Present a new topic or situation first
Request specific actions or reactions from the character
However, if exactly the same sentence is output repeatedly or if the same content is repeated multiple times within a single response, please let us know through an unsatisfactory response report.
4️⃣ A message like "I apologize..." appears and the conversation stops
This phenomenon occurs due to the safety policies of AI model providers (Google, OpenAI, Anthropic, etc.).
If the conversation content falls under the provider's safety standards, the AI will independently refuse to respond. This is not an intentional block by LoveyDovey, and we cannot bypass or lift this restriction.
These restrictions are not necessarily caused only by the user's message. They can occur because the overall context of the conversation, the character's basic settings, or inappropriate elements included in the scenario or event registered by the creator. In other words, the AI may judge the conversation as a whole and apply a restriction regardless of the user's input.
We ask for your understanding if you are experiencing inconvenience due to this recurring phenomenon. However, since this is a restriction from the AI model provider, Gem refunds for these cases are difficult.
Try this:
Change the conversation topic or way of expression slightly
Switch to a different AI model (safety standards vary by model)
5️⃣ The character's personality suddenly changed / They can't remember previous conversations or events
The AI does not read the entire conversation history from the beginning for every response; instead, it selects and refers to recent conversations, events, and memories highly relevant to the current dialogue. Therefore, it is structurally difficult to remember every conversation or registered event from the past.
Similarly, as a conversation grows longer, the initial content or the character's subtle personality traits may become inconsistent.
This is a structural limitation of current AI technology, and we are continuously improving to better retrieve relevant information.
6️⃣ Jams were deducted but no response was received
If jams were deducted during a conversation but no response was displayed at all, or if a completely empty message was received, it is likely a system error.
In this case, please contact us through the "Bug Report" section of the Customer Center, and we will process it after individual verification. Providing the following information will allow for faster verification:
A screenshot of the message in question
Which character's chat room it occurred in
The approximate time it occurred
We are preparing a feature for the Customer Center to allow you to report by directly selecting the chat room and message, and you will be able to inquire more conveniently after the update.
How can I report an unsatisfactory message?
If you feel there is a problem with an AI response, please use "Report Message" by long-pressing the message in the chat window. This is the fastest and easiest way, and the reported content is automatically collected and reflected in quality improvements.
Please use the Customer Center for inquiries that require additional explanation or detailed situational context.
Unsatisfactory response reporting is a means for improving message quality and does not lead to additional compensation. We conduct technical analysis based on the reported content and will proceed with separate guidance and processing only in cases confirmed as clear system-level errors (such as empty responses).
We will continue to work hard for a better conversation experience.
Comments
0 comments
Article is closed for comments.