AI Training, Service Tools & Prompt Engineering (PDF)

Nov 27^th^ -- train the trainors Real AI Today How humans interact with AI Talk to Caleb about it maybe? Prompt engineering: get the most out of Genrative AI and language models - Token: small unit words are broken into - Large Language model: trained to predict the next token - Based o...

Nov 27^th^ -- train the trainors Real AI Today How humans interact with AI Talk to Caleb about it maybe? Prompt engineering: get the most out of Genrative AI and language models - Token: small unit words are broken into - Large Language model: trained to predict the next token - Based on transformer architecture "attention is all you need" by google researchers, 2017 - Reinforcement learning from human feedback (RLHF) - Prompt: the text you input into Chat GPT - Instructions, Questions, input data (character), - Use AI to give you a summary - C.R.E.A.T.E. - Character: who is it writing as? - Request - What do we want it to do? Include context, specific information, details - Examples -- Add examples for tone of voice, similar information - Adjustments: don't use bullet points, use subheadings to break up text, etc. - Type of Output: How long? Bullet points? Email? - Extras: ignore everything before this prompt -- a way to start over, Ask me questions before you answer, Explain your thinking - Use AI to make your work better, not more or faster - Think of AI as a collaborator instead of search engine - **DOUBLE CHECK ACCURACY** - You should know what you want the output to look like - **Datasite AI Tools for Service** Datasite has released AI tools for the Service team to make your job easier. - Service Replies in Chat: this gives you AI suggestions to answer to your client's query. If you find one is a good answer, you can just click on Post to send it directly to the client with no changes or click Edit if you would rather make changes before sending the message. - Einstein Copilot: This AI can answer your questions and help you find the right knowledge article. - Einstein Email Summaries: This gives you a Summary of all the emails in a case. This does have a limit of XXXXXXXXX characters - Knowledge article writing - Please remember to read the whole response before sending to a client, AI is not guaranteed to be accurate. **[List of Actions used in Lana Service Agent]** **Find Similar Opportunities:** Given the ID of an opportunity, identifies a list of opportunities that have similar qualities (such as field values or related records), but aren\'t duplicates of the input opportunity. For example, \'Find me opportunities similar to this one.\' \'What other deals have we won?\' or \'Show me other deals in the same industry.\' **Answer Questions with Knowledge:** Answers questions about company policies and procedures, troubleshooting steps, or product information. For example: "What is your return policy?" "How do I fix an issue?" or "What features does a product have?" **Draft or Revise Email:** Creates an email draft or revises the latest generated email based on user input. **Identify Object by Name:** Finds the Salesforce object API name by extracting an object name from the user input. Identify Record by Name: Searches for Salesforce CRM records by name and returns a list of matching Salesforce CRM record IDs. **Query Records (Beta):** Finds and retrieves Salesforce CRM records based on user input and specific conditions, such as field values. This action automatically identifies the correct records and object type. e.g. \'Find all open opportunities created in the last 5 days sorted by created date.\' **Query Records with Aggregate (Beta):** Answers aggregation questions (such as count, sum, max, min, or average) about Salesforce CRM data based on the user input and specific conditions, such as field values. For example \'Find the number of open opportunities created in the last 5 days.\' **Summarize Record:** Summarizes a single Salesforce CRM record. You must call summarizeRecord only if the user explicitly asks for a summary (e.g: \'Summary\', \'Recap\', \'Highlights\'). This action should be called only when there isn\'t a more specific summarization action.