Alright, so now you know about the different types of prompt design. Zero shot, one shot and few shot prompting are techniques that can be used to get better or faster results from a large language model like GPT:

• Zero shot prompting is where a model makes predictions without being given any examples for training
• One shot prompting is where a model is given a single example for training
• Few shot prompting uses a small number of examples for training (usually between two and five examples)

(If you want to dive deeper on understanding these, I recommend reading this resource.)

Now it's time for you to have some real fun and get your hands a bit dirty! For this exercise follow the below steps:

STEP 1 (GPT-3):

Think of a generative AI problem you want to solve with prompt design + LLMs, eg. a compelling idea for a ghost story or a succinct and eye-catching tweet for a news item.

Now I want you to go into the GPT-3 Playground and try to solve the problem with the three methods we've learned:

(i) First try to solve the problem with zero shot prompt design,
(ii) Second try to solve the problem with one shot prompt design,
(iii) Finally try to solve the problem with few shot prompt design.

Each time observe the results and think about which method produces the best quality completions for your problem, and what the trade-offs are.

STEP 2 (ChatGPT):

Remember how ChatGPT keeps track of the conversational context, so you can use multi-step prompts ("prompt chaining") to get better results?

Now I want you to think of a generative AI problem you want to solve, eg. reviewing an apology message to ensure it sounds genuine or inventing a new sport, then use ChatGPT to incrementally improve the response.

Hint: messages giving additional context, eg. "Well, the sport needs to be playable in a gym and can't rely on expensive equipment" or offering feedback, eg. "Something a little less sentimental would be best." will be handy here.