Imagine that you are teaching a friend how to do something they’ve never tried before. Naturally, you would do this by give them a some examples to make it easier for them to understand what they are being taught. This analog is how I can explain to you how GPT can learn from your prompts based on examples.
You can steer ChatGPT towards the kind of answers you’re looking for just be giving it examples and using prompt techniques.
Here are some easy and entertaining examples to see this in action:
- This example is being used on GPT to convert temperature from Celsius into Fahrenheit.
- Prompt: “I am going to give you some temperatures in Celsius, and I’d like for you to turn them into Fahrenheit. For example: 0°C is 32°F, and 100°C is 212°F. Now, what is 25°C in Fahrenheit?”
How ChatGPT Learns: It sees your examples and translates this as: ‘I need to convert temperatures like that!’ So, it does the math and tells you, “25°C is 77°F.“
2. This example is being used on GPT to figure out ‘feelings’ in Sentence.
- Prompt: “Let’s play a game with sentences. Tell me if they are happy, sad, or just neutral. For example, ‘I love sunny days!’ is happy, but ‘I’m really disappointed with the service’ is sad. What about ‘The movie was groundbreaking’?”
How ChatGPT Learns: It looks at your examples and understands you are asking about the mood of the sentence. So, it responds, “That sentence sounds happy!”
3. This example is being used to create stories about a Cat.
- Prompt: “I would love a funny story about a talking cat. Something like, a cat named Whiskers wanting to drive, or Bob the cat cracking jokes. Can you make up another one?”
How ChatGPT Learns: It gets the idea you’re looking for a humorous cat story. So, it might spin a tale about a talking cat going on a wild adventure.
4. This example is being used to translate language.
- Prompt: “Can you help me translate some text English into French? For example, ‘Where is the library?’ becomes ‘Où est la bibliothèque?’ Now, how do you say ‘How do I get to the train station?’ in French?”
How ChatGPT Learns: It sees your translation example and understands that you want English phrases turned into French. So, it translates, “Comment puis-je aller à la gare?”
In each of these scenarios, you’re essentially engineering ChatGPT to give you specific outputs by way of examples. I would like to coin this term as GPT Looping: a way of creating a feedback loop between the prompt and the AI’s responses, where the end of one response seamlessly connects to or influences the beginning of the next, creating a continuous, cohesive flow.
I have leveraged this trick by using it in social media content automation. By designing a prompt in parts to generate specific content templates on a loop.
See my blog post on how I iterated this Here :
Leave a Reply