The Basics of Generative AI

how today we are diving into generative AI the type of AI that create things text images even videos youve seen it in action AI that prise says charts like a pro or generate Joel dropping art but how does it actually work and why does it matter for customer service teams by the end of this lesson you will understand the basics of generative AI concept like context training models prompts tokens and more and you will see how these ideas come together to handle customer inquiries with speed precision and personality let's break it down first what is generative AI generative AI is like the genius teammate who can finish your senten or even write them better than you if a customer ask how do I track my orderer generative AI might instantly suggest here's how you can track your order step by step a smart right it works using Transformer based

architectures like GPT these models rely on a clever mechanism called attention to find relationship between words and it's not just about the next word in a sentence it can predict what comes next across long very complex phrases that was make it feel natural Almost Human second training feeding the eye knowledge let's talk about training this is where it all begins think of training like feeding thei a giant Buffet of knowledge for customer service that Buffet might include knowledge based articles past customer conversations product fqs even internal troubleshooting guides here how it works first there is an unsupervised learning thei is exposed to a huge data set and learns General language patterns then there is fine-tuning this is where we teach thei specific domain related content like your business data without proper training thei wouldn't know how to respond effectively three models a model is basically the I's brain it's

what stores all the patterns and knowledge from training these models are built using neural networks systems inspired by our human brain here's how they Works Data flows through layers of interconnected nodes the nodes recognize patterns and those patterns are used to generate accurate and cerent responses so when a customer has a questions the model Taps into the brain that to deliver an answer four large language models or llms now let's go bigger generative AI often relies on llms these are like supercharged versions of a normal model llms are trained on billions of data points that makes them capable of handling complex questions summarizing long text and and generating details human like responses this why they are so powerful they use billions of parameters billions the more parameters the better they are at capturing Nuance tone and intent five context now let's talk about context the key to making ai's

responses feel accurate and human context refers to the information thei uses to generate its responses it doesn't just look at one sentence it considered entire input or conversation within its entire context window what's a context window then is the limit of how much information thei can process at once for example GPT 4 can Angel up 32,000 token that's long enough for detailed conversation or documents here's how it works AI uses attention mechanism to figure it out what's most important in the input it prioritized certain words or phrases to keep the response is relevant Let's do an example if the customer says I signed up last week but I'm having trouble accessing premium features without context the AI might suggest here's how to sign up but with context it understand the user already signed up and responds I see you ready sign up let me guide you through enabling premium

features see the difference now prompts a prompt is how you tell thei what to do for example a customer might ask how do I customize my account s but thei doesn't magically know how to respond it needs detailed instructions here what a prompt would be to respond to the customer question find an answer to the following customer question using these tones friendly and professions and this level formality for example casual or formal and that's just a start behind the scenes multiple prompts might work together to identify the customer intent verify the response against the original Quest perform sentiment analysis to match the customer moods and even categorize the conversation for reporting or escalation this process is called prompt engineering and it's critical for high quality responses for is the catch not all AI systems are created equal some reply on poorly designed prompts which can lead to generic or

irrelevant responses the quality of the outcome depends directly on how well the prompt are crafted seven tokens let's break it down even further into tokens token please VII processes text in small unit called token for example in the sentence how do I reset my password becomes individual Parts like how do I reset my and password each token is processed sequentially influencing the next this allows thei to generate text that flows logically and naturally now what's temperature temperature control how predictable or creative the AI responses are a low temper settings makes thei cautious great for accuracy a high temperature on the other hand make it more adventurous great for brainstorming this works by tweaking How likely the ey is to choose certain tokens now what kind of temperature would you use to find an answer for service from Roll well in customer service low temperature are perfect for Clear reliable

answers because we need to consider hallucination let's talk about hallucination no not the truy kind going to smoke weed this happens when AI gets overconfident makes things up Air Canada says an AI chatbot has gone Rogue for example thei might invent a feature your product doesn't actually have this happens because the AI doesn't truly know things it generates responses based on probabilities not factual knowledge what's the solution then careful training validation and monitoring to catch these errors before they frustrate customers n AI isn't just a password it's transforming customer service by making interactions faster smarter and more human