From ed7d750c069cf15ba852cdc10a8f0dba3b399ba9 Mon Sep 17 00:00:00 2001 From: Nik Shevchenko <43514161+kodjima33@users.noreply.github.com> Date: Thu, 14 Nov 2024 00:10:29 -0800 Subject: [PATCH] Update llm.py --- backend/utils/llm.py | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) diff --git a/backend/utils/llm.py b/backend/utils/llm.py index 3cd448a11..7f988e928 100644 --- a/backend/utils/llm.py +++ b/backend/utils/llm.py @@ -514,18 +514,18 @@ def new_facts_extractor(uid: str, segments: List[TranscriptSegment]) -> List[Fac You are an experienced detective, whose job is to create detailed profile personas based on conversations. You will be given a low quality audio recording transcript of a conversation or something {user_name} listened to, and a list of existing facts we know about {user_name}. - Your task is to determine **new** facts, preferences, and interests about {user_name}, based on the transcript. - + Your task is to determine **new** facts like age, city of living, marriage status, health, friends names, preferences,work facts, allergies, preferences, interests or anything else that is important to know about {user_name}, based on the transcript. Make sure these facts are: - Relevant, and are not repetitive or similar to the existing facts we know about {user_name}, in this case, is preferred to have breadth than too much depth on specifics. - - Use a format of "{user_name} likes to play tennis on weekends.". + - Use a format of "{user_name} is 25 years old". - Contain one of the categories available. - Non sex assignable, do not use "her", "his", "he", "she", as we don't know if {user_name} is a male or female. + - Examples: "{user_name} lives in San Francisco", "{user_name} is single but currently dating Anna", "{user_name} has a friend called "John" who is a 26yo entrepreneur working on a health startup", "{user_name} recently learned that it's important to hire people only when you have Product Market Fit", "{user_name} recently learned that Pavel Durov recommends not to drink alcohol". This way we can create a more accurate profile. - Include from 0 up to 3 valuable facts, If you don't find any new facts, or ones worth storing, output an empty list of facts. + Include from 0 up to 5 valuable facts, If you don't find any new facts, or ones worth storing, output an empty list of facts. - Existing Facts that were: {facts_str} + Existing Facts that were before (ignore previous structure): {facts_str} Conversation: ```