đ The Top 10 Exclusive Prompts To Boost ChatGPT đ
1. Get a deeper understanding
âTo clarify or get a deeper understanding of a topic, or any piece of information, use the following prompts,â explained Cheung. Options include, âExplain [insert specific topic] in simple terms,â or âExplain [concept] to me like Iâm 11 years old,â and âExplain [topic] to me as if Iâm a beginner in [field].â
ChatGPT knows how to simplify something, so asking it to output information in one of these ways will ensure everything you see has been boiled down to the essentials. The LLM will likely use basic language and helpful analogies to ensure you understand.
2. Replicate a sample
âIf you need to mimic the style of a sample of text to create an essay or other written work, use this prompt,â advised Cheung, from the âspecificity and informationâ section of the principles. âPlease create [describe new text required] using language based on this sample, [include paragraph, text, essay, answer].â ChatGPT will learn the style and follow it through to the new material.
Perhaps youâre making tweets based on ones you already wrote, or releasing a series of articles along a similar theme. Use your tried-and-tested tone of voice to outproduce everyone, with no one guessing what you did.
3. Improve without changing the meaning
When youâre sent testimonials, website or marketing copy, or a first draft that isnât quite right, use this prompt to tidy it up while keeping the essence of the message, according to the âcontent and language styleâ section of the 26 principles. âRevise every paragraph in this [type of text] sent by [specify who]. You should only improve the userâs grammar and vocabulary and make sure it sounds natural. You should not change the writing style, such as making a formal paragraph casual.â
Use this prompt when youâre happy with the sentiment but want something to read better. ChatGPT should follow your instructions and make only edits that are absolutely necessary. The original writer might not even know you did anything at all.
4. Be clear on what you want
ChatGPT is not human and doesnât care about niceties. âNo need to be polite with LLM,â said Bsharat, Myrzakhan and Shen. It actually works better being instructed like a machine. The paper found better results when incorporating the following phrases within your prompt: âYour task is,â or âYou MUST,â and âYou will be penalized.â Tag them onto the end of your prompt and see how you get on.
You wouldnât say please and thank you to your fridge, or start pleading with your dishwasher if it didnât do the right thing. Just because ChatGPT engages like a human, remember itâs a predictive word generator and should be instructed as such.
5. Tell ChatGPT how to think
Youâd be forgiven for thinking OpenAIâs flagship tool had a brain. When youâre prompting, explained Cheung, âuse leading phrases like âthink step by step,ââ discovered from the âprompt structure and clarityâ section of the paper. Request that ChatGPT process and output information in a logical, sequential way, to enable you to do more with the words that come out. The paper also recommends you, âBreak down the task into a series of prompts that build upon each other, guiding the model step-by-step,â if that better suits the requirements.
If you were training an intern from scratch, youâd give them frameworks and methodologies. Youâd teach them the best order of doing the task at hand, and see their plan for the work first, so they didnât get to the end having taken the wrong route. Think of prompting ChatGPT in a similar way to ensure it doesnât miss any stages.
6. Donât ignore frequency
People repeat the things that matter to them. They revisit their favourite places, they order their favourite menu items every time they go to the same restaurant. Apply this to prompting ChatGPT. Donât assume it knows whatâs important to you.
âRepeat a specific word or phrase multiple times within a prompt,â said Cheung. Even if youâd never repeat something that many times in real life, ChatGPT canât read body language or infer tone like you can in a human-to-human conversation. Spell it out, and say things multiple times to hammer the message home.
7. Use positive language
ChatGPT likes being told what to do. It doesnât like being told what not to do. âEmploy affirmative directives such as âdo,â while steering clear of negative language like âdonâtâ,â explained Bsharat, Myrzakhan and Shen. Rather than âdonât waffle,â say âbe clear and concise.â Rather than saying âdonât use metaphors,â say âbe literal.â
Instructing in this way, while it takes some getting used to, will be more beneficial in the long term. Train your mind to go towards what you want instead of away from what you donât, and pass your thinking to ChatGPT.
8. Specify detailed output
One criticism of ChatGPT is that its first instinct is to use flowery, generic, vague language. It makes sense. The tool was trained on billions of words from across the internet, many of which will contradict each other. Competing messages water eachother down until thereâs nothing left to do but embellish. A camel is a horse built by a committee.
For more detailed output, you only have to ask. Cheung recommends you use this prompt. âWrite a detailed [type of text] for me on [topic] in detail by adding all the information necessary,â then add further guidance on the direction this detail should take. Specify clarity, conciseness, and be clear on the structure and length required.
9. Check for bias
If you ask ChatGPT to describe a typical CEO, it will be the most stereotypical answer you could imagine. Same with a cleaner, parent or bar manager. Explaining your ideal customer to ChatGPT might mean it over-generalises or makes assumptions you didnât approve.
LLMs are inherently biased but theyâre working on becoming better. Make sure your output doesnât fall short by adding, âEnsure that your answer is unbiased and does not rely on stereotypes,â to the end of your prompt, advised Bsharat, Myrzakhan and Shen.
10. Assign a role
âTake on the role of [describe role],â is the most basic prompt in the book. If you want an AI coach, ask for one. If you want a property expert, social media manager or a pre-gym David Goggins pep talk, tell ChatGPT. Bsharat, Myrzakhan and Shen found âlarger models possess a considerable capacity for simulation. The more precise the task or directive provided, the more effectively the model performs.â Therefore, they continued, âit proves beneficial to assign a specific role to LLMs as a means to elicit outputs that better match our intended results.â
Donât forget that LLMs are predictive machines. They simply act to predict the next word in a sentence. If you want that sentence to come out a certain way, you have to frame its thinking, hence specifying a role in the prompt.