Skip to main content

Deeper Tuning and Advanced Techniques

Deeper Tuning and Advanced Techniques for Optimizing Your AI (for demanding and technically savvy users)​

This section is intended for users who want to go beyond the standard setup and achieve even more precise behavior from their AI instance. These techniques require a deeper understanding of how BuddyPro works and more experimentation. Before you dive in, make sure you have the basics firmly mastered -- especially creating a high-quality System Prompt and preparing relevant know-how.

Precise Manual Edits of Expert Sub-Role Content (definitions)​

As mentioned in sections 10.6 and 15.3, you have the option to manually edit the content of Google Docs containing the definitions of individual expert sub-roles, which BuddyPro automatically generates based on your know-how in the ROLES folder on your Google Drive.

When can this advanced technique be useful?

  • If even after carefully fine-tuning the System Prompt and adding relevant know-how, you still observe that within a specific, narrowly defined expert sub-role the AI repeatedly fails, gives suboptimal advice, or behaves in a way that contradicts your intent for that specialization. (To identify the role, use the /lastRole command, and for deeper analysis of a response use /investigateAnswer:{message_you_want_to_investigate}.)
  • If you want to insert very explicit directives, instructions, or links into the definition of a specific role that the AI should prioritize within the context of that role. For example:

How to do it (with great caution and after careful consideration):

  • Identify the specific Google Doc with the role definition you want to edit (e.g. Role_FacebookAdsBeginners.gdoc). Role names can also be found in the file ROLES/list_of_roles.gdoc.
  • Create a backup of this file before you start making edits!
  • Carefully review the existing, automatically generated content of the role to understand its structure and logic.
  • Carefully make your text edits. You can add new paragraphs, refine existing wording, or add specific instructions. Try to make your edits as clear, specific, and consistent as possible with the rest of the role definition and with your global System Prompt.
  • Save the changes in the Google Doc.
  • For your manual edits to take effect in the AI's behavior, you must run the update command for this specific role in Telegram, in the chat with your BuddyPro instance:
/updateRole:role_name_without_extension

Warning: Manual edits to role definitions are truly an advanced expert technique. Unskilled or overly extensive changes can paradoxically lead to worse consistency or functionality of the AI. Pavel Riha, for example, states that 99.9% of roles in his Online Strategist AI instance are generated automatically and did not require manual edits. If you are unsure, focus instead on optimizing the System Prompt and the quality of your know-how in sources. Most issues can be solved through these standard approaches.

Changing Your AI Voice for Audio Replies​

Your BuddyPro instance can respond to users not only with text, but also with a synthetic voice (for example, if a user sends a voice message). You can choose from several available voices.

  • To set or change the voice, use the setting in the Dashboard under the Configuration section:
  • How do I find the available voices? The current list of available voices (including male and female variants) and their samples can be found in the official OpenAI documentation. You can also try asking your BuddyPro instance directly about available voices, but the most reliable source is the documentation linked above.
  • This feature adds another level of personalization and can improve the user experience, especially for those who prefer listening over reading.

Voice Options -- Teach Your Buddy to Speak in Your Own Voice​

NEW FEATURE

You can now teach your Buddy to speak with your own voice.

Generating voice messages in your own voice is approximately 2x more expensive than generating messages using standard voices. We estimate that this will increase your Buddy's total costs by about 10--20% (because not all users always use voice messages).

  • To create your own voice, use the command:
/createVoiceClone
  • To test the created voice on yourself, use the command:
/useVoiceCloneForMe:true
  • To activate the created voice for all users, use the command:
/useVoiceCloneForEverybody:true

Experimenting with Advanced Diagnostic Commands​

For deep insight into your AI's "thinking" and for detailed fine-tuning, you have access to a set of diagnostic commands. Using them requires patience and analytical skills.

  • /investigateAnswer:{message}: Responds to the given {message} and also shows you relevant texts from your know-how and the role content that were used to generate the answer. A great tool to check whether the AI is drawing from the correct sources.
  • /lastRole: Returns the name (ID) of the role that was used to generate the AI's last response. Useful for identifying the role that may need manual editing.

By analyzing the outputs of these commands, you can identify very subtle nuances in the AI's behavior and get ideas for even more precise optimization of the System Prompt, the structure of your know-how, or the definitions of expert sub-roles. However, this is an area that requires significant patience and a willingness to experiment. Always proceed carefully, and ideally test changes first on a test profile.