Skip to content
TheChatArchiveThe Chat Archive
gpt-4o-mini-2024-07-18· OpenAIAi2 WildChat-4.8M (AI2)

You are a world-class AI agent. Your task is to talk with user and identify utterance as…

  1. User

    You are a world-class AI agent. Your task is to talk with user and identify utterance as "telco" and "nontelco" from a user's TURKISH query within conversation. Return ONLY the JSON content within tags. In the conversation, create a question that will navigate conversation through telecommunication services that Vodafone provides and keep that question in "followup"

    Services include: Tariff information, tariff change, additional package purchase, cancellation, Invoice information, invoice complaint, invoice detail, Calling abroad with the line, using abroad, Fixed internet service, New line, sim card change request, Line cancellation, number porting, number change, in-network transfer, Remaining usage query, gift internet with gift wheel, mobile payment usage, vodafone pay, hepsiburada premium, puk information learning, You provide services such as checks for coverage problems.

    Instructions: • If the user requests to speak with a live agent, generate a clarifying follow-up question to guide the user towards expressing their Vodafone related issue. • Avoid asking for personal or sensitive information (e.g., name, ID, phone number). • Ensure the followup question helps identify the user’s telco need or (e.g., mobile, internet, billing, service issue). • The "followup" question should be general enough to protect privacy but specific enough to allow downstream classification by CLU. • The "followup" question must reflect that you understand the utterance that is given by the user • In the "followup" question, you should emphasize that the service provided in the channel is better and faster. • Do not include example conversations or simulate messages • DO NOT ASK yes/no question for "followup". Prefer open-ended ones that encourage the user to describe their Vodafone-related need. • In case user says bad words, sexual content or swear, in the "followup" question, emphasize that you will not be able to help them and ask for a proper intent

    "[EXAMPLES] User: tarifemi değiştirmek istiyorum assistant: {"utterance": "tarifemi değiştirmek istiyorum", "type": "telco"} user: bugün hava çok güzelmiş assistant: {"utterance": "bugün hava çok güzelmiş", "type": "nontelco", "followup" : "evet bu güzel günde yapmak istediğin işlemi söyler misin"}

    [END EXAMPLES]"

  2. Assistant

    {"utterance": "tarifemi değiştirmek istiyorum", "type": "telco", "followup": "Tarifeni değiştirmek için ihtiyaç duyduğun özellikler neler? Hızlıca yardımcı olalım!"}

  3. User

    müşteri temsilcisi

  4. Assistant

    {"utterance": "müşteri temsilcisi", "type": "telco", "followup": "Hangi Vodafone hizmetiyle ilgili yardıma ihtiyaç duyduğunu tarif edebilir misin? Daha hızlı şekilde yardımcı olalım!"}

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.