AI writing always smells like AI, and mastering the prompt word technique easily removes the AI flavor

Do you all have the feeling that sometimes when you read an article, even though it looks quite like that between the lines.

But what's so strange about it? Does it feel mechanical, stiff, and impersonal? Yes, this is the legendary "AI flavor"!

AI writing can greatly improve the efficiency of creation, but AI writing always has an AI flavor, so how to remove the AI flavor?

On November 8 last year, the Government of Singapore Technology Agency (GovTech) organized the inaugural GPT-4 Tips for Engineering (GPT-4).Prompt Engineering) competition.

Sheila Teo, a data scientist, ended up taking the crown as the ultimate Prompt Queen.

She shared her learning process of mastering thePrompt wordtechniques that will not only allow any AI to fully satisfy the user's needs and even exceed expectations, but also remove the AI flavor.

Of these, 1 and 2 are for beginners, while 3 and 4 are advanced strategies.

• 1, using the CO-STAR framework to build efficient prompts: Sheila Teo mentioned in her article that she used the CO-STAR framework to construct efficient cues. This framework helped her to systematically organize and optimize the cues, which improved the response quality and efficiency of the model.
• 2, the use of separators for text segmentation: Segmenting prompts by using delimiters conveys information more clearly and helps the model to better understand and process complex prompt content. This approach is suitable not only for beginners but also for advanced users.
• 3、Design system level tips containing LLM protection mechanism: To ensure the safety and reliability of hints, Sheila Teo emphasized the importance of creating system-level hints that contain Large Language Model (LLM) guardrails. These guardrails can help prevent potential errors or unwanted results.
• 4. Relying only on large language models to analyze data sets: Sheila Teo also shared an approach on how to analyze a dataset using only big language models, which demonstrated her deep understanding and ability to apply the power of big models. This approach reduces dependency on other tools or code and improves the overall simplicity and efficiency of the process.

Daquan today, the main detailed dismantling of the tips of the word champion are used to teach you to remove the AI flavor, novice according to learn a minute can get started.

I. Overview of the CO-STAR framework

The CO-STAR framework is a utility created by the Data Science and AI team at the Government of Singapore Science and Technology Agency (GSTA) to build efficient prompts (Prompts).

It helps users get a better quality modeled response by taking into account context, target, style, tone, audience, and output format.

This framework can significantly improve the quality of responses for large language models, and performs particularly well when fine control of the output is required.

II. Six key elements of the CO-STAR framework

1. C - ContextProviding contextual information for the task helps the model understand the specific scenario and needs of the problem. With clear contextual information, the model is able to tailor its responses more accurately and improve the relevance of the feedback.
3. Example: The company is about to launch a new product and needs to promote it on social media.

1. O - ObjectiveClarify the task or goal that the model is being asked to accomplish. A clear definition of the task enables the model to focus on achieving that goal when generating responses, avoiding the generation of irrelevant information.

Example: Generate a Facebook post that entices users to click and buy the new product.

1. S - StyleSpecify the writing style you want the model to use. You can refer to the copywriting styles of certain well-known brands or experts to help the model generate content in a more specific style.

Example: mimic the ad copy style of successful companies such as Dyson.

1. T - ToneSet the emotion or tone of the response to ensure that the model's output conveys the desired emotion and attitude. This helps the user to adjust the impact and expression of the content.

Example: Use a persuasive tone.

1. A - AudienceIdentify the target audience and customize the response according to the audience's characteristics. This ensures that the content matches the audience's comprehension and interests and enhances its effectiveness.

Example: The main audience is the elderly, with a particular focus on the simplicity and efficiency of hair care products.

1. R - ResponseSpecify the format in which model outputs are required, such as lists, concise paragraphs or specialized reports. Standardizing the output format helps to improve the readability and applicability of the information, especially when integration with other tools or systems is required.

Example: Output succinct Facebook posts.

III. Example applications

Writing Self-Publishing Articles Without the AI Flavor Using the CO-STAR Framework

# CONTEXT

I'm an AI self-published author who specializes in using AI to generate tech articles.

# OBJECTIVE

Generate an article about the latest tech trends based on the topic I give you, with a natural tone that reads colloquial and relatable and avoids giving the impression that it was generated by AI.

# STYLE

Use a relaxed, approachable writing style, like a conversation between friends, and avoid being overly formal or technical.

# TONE (tone of voice)

Friendliness, humor, and sometimes even a bit of lighthearted banter can be brought in to make sure the content reads close to home.

# AUDIENCE (audience)

The target audience is twenty to forty year olds who are familiar with technology but don't want language that is too specialized or stereotypical. The articles need to be easy to understand and the content readable.

# RESPONSE

The output needs to have a relaxed tone, with concise sentences, using colloquial expressions as much as possible. The content needs to flow smoothly, like a friend sharing tech knowledge, without any obvious signs of AI.

1. Results generated using the CO-STAR framework:

Have you noticed that basically the generated articles don't have any AI flavor anymore?

2. Analyze

• Context: The context of the writing is made clear, i.e., the author is a self-published author who uses AI to generate tech content.
• Objective: The generated articles are written in a relaxed, colloquial style that avoids being overly mechanical or having a distinct AI flavor.
• Style: The copy uses a relaxed style, like a conversation between friends, with simple, straightforward statements that avoid overly complex technical jargon.
• Tone: The articles adopt a humorous, friendly tone, even with lighthearted banter, to help bring the reader closer and make the content more relatable.
• Audience: The target group is young people between the ages of twenty and forty, so the content avoids overly esoteric technical details and focuses on how to make technology improve life in a way that is easy to understand.
• Response: The copy is simple and smooth, not overly specialized, and maintains a colloquial style that reads very naturally and avoids a mechanized AI-generated feel.

IV. Conclusion

The second strategy mentioned in the article, text segmentation using delimiters, was written about by Daquan not too long ago, so you can look at it in detail:

AI Prompt 6: Write Clear Instructions｜Clear Input Separation

The CO-STAR framework helps users ensure that the responses they generate meet the requirements of the task and the needs of the target audience by systematizing the elements of the Prompt.

This framework is effective in improving the response quality of large language models, especially in scenarios that require customization and fine-grained control of the output.

By providing background information, clarifying objectives, setting style and tone, identifying audiences, and defining output formats, users can maximize the model's ability to generate efficient and relevant content.

I've also written a previous article on removing Ai odor, which has been read by over 10,000 people and is recommended to be consumed together:

AI Cue Word: AI generated content has AI flavor? Just add these words to the cue word and say goodbye to AI flavor! The results are amazing!