🔹 What can these avatars be used for?
These avatars are ideal for research in role-plays, simulations, training, and human-AI interaction. They excel particularly in studies on communication and conversation, language learning, and personalized education.

🔹Why isn't the avatar loading?
In most cases, you just need to wait a little longer and keep the browser window open.
When launching an avatar (e.g., from the avatar library), approximately 200 MB of data must be downloaded.
Depending on your internet connection, this process can take anywhere from 30 seconds to about 5 minutes

🔹Do I need an account?
You need an account to create avatars and to gather log data for experiments.
This is necessary to store data of your avatar and the user logfiles.
Without an account, you can still try one of the avatars from the avatar library.

🔹 Where can I access the log data?
You can download the recorded log data if you are a logged in user by visiting https://avatar-research.com/logfiles/ and clicking the Download button. To properly format the data, open the **.csv file** in Microsoft Word and use the comma delimiter to separate values.

🔹 Is this software free and can I use it?
Yes, it's free and you can use it for research and teaching. If you use it for research, please cite one of the papers in which we presented the software briefly:
Fink, M. C., Robinson, S. A., & Ertl, B. (2024). AI-based avatars are changing the way we learn and teach: benefits and challenges. Frontiers in Education(9), 1416307. https://doi.org/10.3389/feduc.2024.1416307
Fink, M. C., Walter, L., Eska, B., & Ertl, B. (2024). An authoring tool for individual and collaborative learning scenarios with AI-based avatars [avatar-research.com]. EdArXiv. https://doi.org/10.35542/osf.io/ckd9e

🔹 Why does the voice of the avatar stutter?
Stuttering mainly seems to occur a) when using Firefox as a browser b) with bad internet connection. Please use Chrome, Edge, or Safari for better audio quality

🔹 Why are the avatar’s responses too short?
Try increasing the response length by setting the token limit to 150 or 200 tokens.
You can include this in the prompt:
"Each response should be at least 150/200 tokens long. If answers remain too short, gradually increase the token limit."

🔹 Why does the avatar’s tone or style keep changing?
Ensure the avatar has a well-defined prompt that clearly specifies its personality, tone, and response style.
Ideally, create a detailed prompt—similar to a multi-page actor’s script—that outlines behavior, speech patterns, and key traits.
A vague prompt results in inconsistent responses; even the best actors need a strong script.

🔹 How do I stop the avatar from using emojis, LaTeX formulas, or bullet points?
Explicitly state these restrictions in the prompt.
Example:"The AI should not use LaTeX formulas in this math avatar. Responses should be in natural language, suitable for synthetic speech output."

🔹 Why does the avatar forget the conversation?
Increase the memory token limit. A good starting point is 400 tokens to improve recall and conversation continuity.
For more complex conversations, a higher limit of 800 or 2000 tokens might improve recall further.

🔹 How can I make the avatar more knowledgeable on a specific topic?
Provide additional background information in the prompt.
You can include publicly available data (e.g., research abstracts, textbook excerpts, or AI-generated summaries).
However, ensure the content is detailed enough to guide accurate responses.