Skip to main content

Learning Methods in AI: Zero-Shot, Few-Shot, and Many-Shot Learning

Lesson 4 from: Getting Started with AI Prompt Engineering

Mark Hinkle

Learning Methods in AI: Zero-Shot, Few-Shot, and Many-Shot Learning

Lesson 4 from: Getting Started with AI Prompt Engineering

Mark Hinkle

buy this class

$00

$00
Sale Ends Soon!

starting under

$13/month*

Unlock this classplus 2200+ more >

Lesson Info

4. Learning Methods in AI: Zero-Shot, Few-Shot, and Many-Shot Learning

This lesson breaks down how models like ChatGPT make predictions based on varying amounts of task-specific examples during training, from no examples (Zero-Shot) to several instances (Many-Shot).

Lesson Info

Learning Methods in AI: Zero-Shot, Few-Shot, and Many-Shot Learning

1 Let's talk about zero, one, and many shot prompts. 2 So zero shot prompts are when the model makes decisions 3 based on the training data it has, 4 but it doesn't have an example. 5 So our simple prompt that we provided in the earlier example 6 was a zero shot learning prompt. 7 However, you could provide an example 8 of what you wanted the LinkedIn post to look like, 9 or if you did many shots, 10 you could actually create many examples. 11 So let's just look at one. 12 We're gonna open my prompt example here. 13 Here's an example of a zero shot prompt. 14 "Create a compelling description 15 for a Fiverr gig offering graphic design services." 16 Now I'm gonna paste that in, 17 and I'm gonna hit enter. 18 In the output, it decided to come up 19 with a transformative graphic design services 20 for your brand. 21 And a description 22 and a description of how I'm committed to excellence. 23 And notice ChatGPT even puts a disclaimer 24 that says "This is a mock description 25 and may n...

ot be based on your actual skills and experience." 26 So it knows it doesn't have much information. 27 Now let's go back and do a few shot prompt. 28 And I have two examples 29 of what I want for the Fiverr description. 30 I'm gonna do GPT-4, 31 I'm gonna make sure I have my plugins changed in 32 'cause it could pull some information from the internet. 33 Okay, so now I've created a few shot. 34 It has two examples 35 and it's gonna echo basically the responses 36 that I had based on my two examples in the prompt. 37 And it's a compelling description. 38 It's not a full listing for a Fiverr gig, 39 but it's a compelling description. 40 And because it knows from the 41 instructions earlier that I do consulting, it got confused. 42 So I'm gonna take that out of my instructions. 43 So I have no instructions. 44 I'm gonna save that. 45 I'm gonna try this again and see if it does. 46 So if you see logo design services, 47 social media management, 48 let's see if it gives me the same thing. 49 And it did. 50 And the reason it did was 51 because it remembers our conversation. 52 And so this is like a conversation between two people 53 where they remember all the details earlier 54 so you don't have to reference them all the time. 55 So if I want a different result, what I would do 56 is I'd go to a new chat. 57 I've cleared my instructions, 58 so my instructions are clear now 59 and I'm gonna post it in 60 and we'll see if we get a different result here. 61 It created logo design services, social media 62 and management services. 63 It got a little closer. 64 What I could do is say, 65 define this and refine it a little bit 66 and say, since the two examples were logo 67 and design, let's just say we are going 68 to create a Fiverr gig 69 description for logo design services. 70 Notice that because now I've sort of iterated on the prompt 71 by saying create a Fiverr gig description 72 for logo and design services, 73 it actually says, "Welcome to your ultimate 74 destination for logo design." 75 That iteration is what I was talking about. 76 You can iterate on the first system prompt like I showed you 77 a notion or you can iterate as you go. 78 All right, that brings me to my next topic. 79 Chain of thought prompts. 80 So chain of thought prompts are a series 81 of intermediate reasoning steps 82 and it improves the ability of the language model 83 to understand your train of thought. 84 It actually is designed to show the AI how you're reasoning 85 through a problem or coming to a conclusion. 86 Now we understand how we format prompts. 87 We understand a few shot, many shot, 88 zero shot learning chain of thought. 89 Let's get to the good stuff, let's start mastering prompts.

Class Materials

CLASS MATERIALS

Super Prompts

Ratings and Reviews

Mayoon Boonyarat
 

It's a really good starting point and gives you an overview. The CreativeLive team is really fantastic to bring Mark here. You'll know that the most important skill is the prompt skill. Talk with AI.

Student Work

RELATED ARTICLES

RELATED ARTICLES