How to Use GPT-3 – A Quick Guide
If you’re late to the AI chatbot craze, this article will get you up to speed. You will learn how to avoid common mistakes, the “hidden” limitations on usage, and most importantly, how to prompt the software effectively so that you don’t end up with vague answers or even wrong ones. Here are the most important things you need to know as a first-time user.
How to Use GPT-3
When browsing for GPT-3, it’s important to note that it’s different from ChatGPT and other OpenAI software available today. You are looking for something called GPT Playground. Playground is free to use and requires a quick log-in through Google, Microsoft, or an OpenAI account. The interface is beginner friendly so, once you are in, you can hover over any item with your cursor and a bubble will come up giving you the specifics of that item.
Mind the Tokens
GPT-3 uses tokens to measure the size of your interaction. Be aware that there is a limit on how much text you can generate. Here’s how the tokens work:
- 1-2 sentences are approximately 30 tokens.
- 500 words are approximately 680 tokens
Tokens are priced differently based on the model that you want to use. When you first log in , you’ll receive free trial tokens which will be enough to give the software a go. Token usage is also tracked for you so you can plan how best to spend them.
You can limit token spending by setting the length of any given answer. Both your prompt and the software’s response count toward your token usage. Fortunately, you can find a token counter in the bottom right corner of the screen under the chat box. Note that leaving space between the text also spends tokens.
GPT Playground has a few models which you select under the drop-down menu. Models are featured on the right-hand side of the screen. These determine the quality of the answer as the models have been trained on different data. At the time of writing, the model “text-davinci-003” is the most advanced.
Everything About Prompts
Prompts are your attempts to get GPT-3 to give you a specific response. As such, you should set aside some time to learn how to approach this step. Always keep in mind that Playground doesn’t know exactly what you want when you first type something into the program because it’s context-dependent. To start effectively, provide as much information as possible. Here’s an example of how to do so with a generic prompt:
Generic prompt: Write a story.
Better prompt: Write a short horror story.
Even better prompt: Write a 1,000-word short horror story about a group of friends getting lost on a hiking trip and barely getting out alive.
Let’s take a look at what goes into crafting effective prompts.
Prompting for Style
You can use the GPT-3 database to get a response in a particular style. This can be a genre or style of writing. You can also limit it to a specific book or news article and tell it to form an answer based on that. Moreover, it’s possible to mix an author with a genre in which the author never wrote and a topic they never covered.
Here’s an example of how that could look:
- Write a short sci-fi fable in the style of Ernest Hemingway where the moral of the story is to not be jealous.
Prompting for Word-Count
As mentioned before, your interaction with Playground is limited to the number of tokens you can use. This requires some careful planning on how many words you want to dedicate to each answer. Do this by changing the number under the scale named “Maximum length.” Additionally, follow the token count under the lower right corner of the chat box.
Here are a few prompts:
- Write a full response to this question in under 150 words.
- Write five short questions about this topic.
- Rewrite this text and limit it to 250 words.
- Tell me why this computer code isn’t doing (insert task) but limit your answer to four sentences and a fixed version of my code.
Changing Language-Specific Points in a Text
Sometimes it’s important to simplify a given text. GPT-3 can do just that based on standardized language ability levels as well as specific grammatical constructions. Simply tell it to avoid using the passive voice or to use a lot of modal verbs and it will do so. It can simplify concepts and even clarify and explain legal or medical terms.
After you paste a text into the chat box and see how many tokens it will take up, try the following prompts:
- In the text above, rewrite all professional jargon into plain English.
- Simplify the sentences and rewrite them into B2 English Level without losing the importance of the content and the meaning.
- Explain this math problem to me without using any math jargon.
- Ask me three questions about the text above to check my understanding. Don’t ask me about any information that isn’t in the text. After I answer, tell me if my answer is correct or not, but don’t give me the correct answer until I request it.
Limiting the Dataset
You’ll often have a general question that you can only make specific by limiting it to a certain historical period, a single country, activity, or person. Questions like, “What is the most popular rock band?” require more information about the musical style, the era, etc. The question, “What is the best car?” also needs to include a brand or horsepower, whether it’s gas or electric, etc.
Here are those questions made more specific:
- What is the most popular rock band in Japan formed before 1995?
- What is the fastest German-made car that can be considered a modern classic?
Aside from the manipulation of text, you can use this language model for the following:
- Writing, rewriting, and fixing your code in any programming language based on prompts
- Train it to automate responses to customer questions on websites
- Automating email generation
- Translation work
- Creating images from text prompts
Because GPT-3 is open-source, it means that you can download it and feed it your data. If you’re skilled enough, this will allow you to train it to do things that are specific to your needs.
Be Careful About This
It’s important to note that GPT-3 doesn’t have access to data after September 2021. If you ask it a question beyond that point in time, it may give you an incorrect or outdated answer and even go as far as making something up completely.
Playground is also not suited to answer moral questions or questions about life choices. It’s not a therapist or a doctor. Although it can simulate the answer of a professional – even a specific one, it’s not a thinking agent. It is just, as the name suggests: a playground.
Have you ever used this powerful AI tool in your life? Are you happy with the results? Let us know in the comments section.