The development of artificial intelligence (AI) technology has led to the creation of conversational agents, like ChatGPT. While these agents can generate text based on a given prompt, users may find it challenging to personalize the language to their liking. I’ve been prompting and poking AIs for a while now, and getting the most out of them can be tricky - so I want to share some of my techniques with you, especially around personalizing what is generated.
Use Your Tone - Set the Mood:
One of the most effective ways to make ChatGPT sound like you is to use your tone of voice. When conversing with ChatGPT, speak to it in a way that you would like it to respond. It is a prediction machine, and it uses the context given to it, including the conversation it has with you. So if you want a serious and professional-sounding response, make sure to set that tone in your conversation. On the other hand, if you prefer a more relaxed tone, let ChatGPT know that too. By setting the tone, you can make ChatGPT sound more like you.
Contextualize the Conversation:
The context of the conversation is critical to getting ChatGPT to respond in a way that sounds like you. If you want ChatGPT to use language that feels more natural to you, make sure to provide the necessary context. Use the same language and vocabulary that you would typically use in a conversation with someone. Explain who your audience is. Explain your level of understanding. Provide the necessary context for the specific topic of conversation so that ChatGPT can offer relevant and accurate responses.
Modify the AI-Generated Text:
ChatGPT is a generation tool, not a human. As such, it may not always get things exactly right. One way to personalize ChatGPT's language is to modify the generated text. If a response seems too corporate or contains too many emojis, let ChatGPT know that. Work with it to get the text to your liking. If it doesn't produce the desired result, you can copy the text and edit it yourself. By modifying the text, you can get ChatGPT to generate responses that sound more and more like you as you learn what works and you get better examples to give it.
Give it a Job:
ChatGPT likes being told what to do, and explicitly so. If you tell it the task you want it to perform it will try to do so with single-minded dedication. If it’s something it’s good at, it’s great. If it’s something it sucks at, you’ll regret asking. But either way, you won’t be left wondering and you’ll have gained value from the attempt.
Seriously, this is so powerful - give GPT examples of what you want it to do. Separate sections with weird characters like # or ⌂ and tell it what’s what, like this example:
“Digital friend, your job is to format sentences more persuasively for the purposes of writing a winning factum - make them punchier. Here are some examples of input sentences and their corresponding more persuasive versions:
- The defendant's actions, which were not in line with the standard of care that is typically expected in such circumstances, have resulted in a situation that can only be described as detrimental to the plaintiff.
- With regards to the plaintiff's claim for damages, it should be noted that the amount requested is excessive in light of the fact that similar cases, when examined, have seen lower amounts awarded by the courts.
- There exists an abundance of evidence that points in the direction of the defendant being fully cognizant of the potential negative consequences that their actions might lead to, however, they chose to proceed regardless.
- The defendant's negligence violated the standard of care, causing significant harm to the plaintiff.
- The plaintiff's claim for damages is excessive compared to precedent, warranting a reduced award.
- The defendant knowingly risked negative consequences but proceeded regardless, demonstrating recklessness.
Now, fix these poor versions:
- It should be taken into consideration that the defendant, in this particular situation, failed to adhere to the Occupiers' Liability Act of Ontario, which was established to govern the responsibilities and duties of individuals who are in control of premises, leading to the unfortunate accident that the plaintiff experienced.
- In the province of Ontario, there is a limitation period that has been established by the Limitations Act, 2002, which stipulates that actions of this nature are to be commenced within a two-year time frame; however, in the current case, the plaintiff has not acted in a timely manner and has initiated their lawsuit beyond this specified period.
- According to the Family Law Act of Ontario, the spousal support that is being demanded by the plaintiff is not in line with the calculation guidelines that have been set forth and, as such, the plaintiff's request for the aforementioned support should be subjected to a thorough reevaluation by the court.
AI Assistant, not AI Replacement
It’s important to remember that ChatGPT is there to assist you, not replace you. By working with ChatGPT, you can create text that feels natural and personal. Collaboration between user and AI is key; ChatGPT can generate text based on its training data, but it's up to you to personalize it.
Personalizing ChatGPT’s language can help make it sound more like you. By using your tone, contextualizing the conversation, modifying the text generated, and collaborating with ChatGPT, you can create text that feels personal and natural. Use these tips to get the most out of ChatGPT and achieve your desired outcome.