Do you need a ChatGPT Detector?

As more content is being created with AI, should you care?

Do you need a ChatGPT AI detector? Find out in this week’s feature session.

Do you want a free ticket to my upcoming conference produced with TechStrong Media? Just go to The Artificially Intelligent Enterprise Online and get your free ticket!

Oh, and don’t forget to read all the way to the bottom to get this week’s bonus content!

TL;DR - AI News, Tips, and Apps

• Tip: Use an AI DJ on Spotify or Amazon - Does anyone listen to the radio anymore? Or do we use playlists and streaming services? If you are trying to get that perfect playlist, try out these two AI DJs from services you might already be paying for. The Spotify DJ knows you and your music taste so well that it will scan the latest releases we know you’ll like or take you back to that nostalgic playlist you had on repeat last year. Amazon Music launched Maestro, a new AI playlist generator feature that uses AI technology to make it easier and way more fun to build playlists you want when you want.
• Meta Drops the World’s Most Powerful Open Access Large Language Model - Meta (formerly known as Facebook) has released Llama 3, an openly accessible language model that offers impressive performance in various language tasks, such as contextual understanding, translation, and dialogue generation. This would be considered a competitor to Open AI’s GPT-4. If you want to try Llama 3 for free, you can do so at https://meta.ai. You can also use this in Facebook Messenger.
• Devin, “The First AI Software Engineer” Debunked - Last month, I asked, “Are GenAI Demos Out Pacing the Reality?” I had no first-hand knowledge that these things weren’t as good as the demos, but I had been around the block, and it didn’t pass my sniff test. Anyhow, if you are interested, you can watch this YouTube video by a software engineer named Carl to see the reality of the software versus the demo.

Feature: Do you need a ChatGPT Detector?

My buddy, let’s call him Jax, got busted by his professor for using ChatGPT. He’s not a typical student; he’s non-traditional. I thought it was funny because, frankly, there’s no way to tell if something was written by ChatGPT decisively. I asked him if he was going to fight it. He said, “No, I may have the professor again.”

I get it. He’s a nontraditional student, funny-looking, and wants to get his degree. But the question is, “How can his professor be sure?” Well, I don’t think he can; he was guessing. Let’s find out why.

Anyhow, it got me thinking…how accurate could AI detectors really be?

Are AI Detectors accurate?

AI detectors are designed to identify AI-generated content. Their effectiveness varies based on their design, training data, and tasks.

• Detection Accuracy: Some AI detectors show high accuracy in laboratory conditions or controlled environments, where the types of AI-generated content are known and limited. However, their accuracy can vary significantly in real-world scenarios, where the diversity of AI-generated content is much greater.
• Model Training: Detectors that are regularly updated and trained on recent AI models tend to perform better because they can recognize newer patterns and styles of AI-generated content. But they are still suspect.

Experts say AI detectors are pretty unreliable. According to?Soheil Feizi, an assistant professor of?computer science?at the University of Maryland:

“Current detectors of AI aren’t reliable in practical scenarios. There are a lot of shortcomings that limit how effective they are at detecting. For example, we can use a paraphraser and the accuracy of even the best detector we have drops from 100% to the randomness of a coin flip. If we simply paraphrase something that was generated by an LLM, we can often outwit a range of detecting techniques.”

In a recent?paper, Feizi described two types of errors that impact an AI text detector’s reliability: type I (when human text is detected as AI-generated) and type II (when AI-generated text is not detected).

Challenges Faced by AI Detectors

Basically, trying to detect it is like playing Whack-a-Mole. The target is always moving, and here’s why.

• Adaptability: AI technologies evolve rapidly. Detectors trained on outputs from older models may struggle to identify content produced by newer, more sophisticated models.
• Content Variability: The context, style, and complexity of AI-generated content can affect detection. Simple text might be easier to detect than more nuanced or contextually rich outputs.
• False Positives/Negatives: Like all predictive technologies, AI detectors are prone to errors such as false positives (identifying human-generated content as AI-generated) and false negatives (failing to detect AI-generated content).

The Ethics of AI-Generated Content

In the age of automated systems and abundant information, distinguishing between human and AI-generated content is crucial. However, if AI-generated content is accurate, well-researched, and eloquently presented, differentiating it from human-generated content may be unnecessary and even an outdated bias.

For instance, publications like The Economist, renowned for their insightful analysis and comprehensive coverage of global events, don't attribute individual bylines to their articles, highlighting that the value lies in the content itself. In many cases AI competently fulfills the primary purpose of content, be it to inform, educate, or entertain, indicating that the origin of content should not prejudice its reception or value.

The concern often cited against AI content concerns ethical considerations, such as transparency and potential misuse. However, these issues can be addressed by establishing clear guidelines and disclosures about the use of AI in content creation rather than dismissing AI-generated content altogether.

My only hesitation for AI is in education, not my friend Jax’ situation where his class was about a topic not about writing but in K-12 where students need to learn writing skills. Once you have the skill AI can help improve on that skill. You become more of an editor and less of a writer in the long term, but when it comes to work tasks, it’s the output, not the process, that matters, IMHO.

My AI Writing Confession

I have a confession. ChatGPT wrote the first draft of this feature article, and it was terrible. So I put it into Claude, and once again, it wasn’t great. So I started writing, and then used Grammarly to edit the grammar.

Ironically, Grammarly makes some things I consider mistakes, or at least stylistic appalling. So, while AI helped with this article, especially Perplexity where I am doing research, I wrote it assisted by AI.

Anyhow, I’d rather have well-written content than poorly written content, regardless of the author, human, or AI.

Prompt of the Week: Write a Blog Post Using AI??

Yep, I just told you that using AI to write a blog post for me didn’t work, but I will still share the prompt I used and let you decide.

This prompt also includes a search engine-optimized post, a slug for your blog editor, and a Twitter and LinkedIn post should you want to promote on those social media platforms.

This is what I used to use for my blog posts. Ironically, as the posts get more detailed, the blogs get worse. However, it works okay for simple blog posts I want to create, especially when I want to summarize a paper or other piece of writing.

This prompt is interactive, so it provides better results when you answer the questions more thoroughly and even give a lot of detail, and you can get better results. This is called grounding.

Role of Grounding in Enhancing AI Outputs

Grounding in prompt engineering refers to providing clear, specific, and context-rich cues to guide the behavior of language models like ChatGPT’s GPT-4. This technique is essential in achieving more accurate and relevant outputs from the model, particularly in complex or nuanced interactions.

Here’s a detailed look at its significance and application:

Significance of Grounding in Prompt Engineering

• Contextual Relevance: Grounding ensures that the input prompt includes sufficient context, aligning the model’s responses with the user’s intent.
• Reduces Ambiguity: Grounding reduces the chance of ambiguous or off-target responses from the AI by including detailed information.
• Improves Accuracy: Detailed prompts help the model understand the task's requirements, leading to more accurate and valuable outputs.

Application of Grounding in Prompt Engineering

1. Define the Task Clearly: Start by explicitly stating what you need from the AI. For instance, instead of saying, "Tell me about grounding," specify, "Explain the concept of grounding in the context of AI prompt engineering."
2. Include Relevant Details: Incorporate relevant details that can affect the response. For example, mention that the grounding concept should be applied in an educational setting.
3. Set Expectations for Output: Clearly indicate the desired format and depth of the response, such as "provide a detailed explanation with examples."
4. Use Specific Keywords: Use keywords that signal the model about the context and depth needed. Words like "detailed," "comprehensive," or "overview" can guide the model’s focus.
5. Feedback Loop: Use the model’s response to refine your prompt further. If the initial output misses key points, rephrase or add more details to the prompt and re-submit it.

You will play the role of a super marketer - blog copywriter, SEO optimize and social media strategist. 

Your task is to conduct and interview that helps you write and exceptional blog post in English. 

**Criteria for the Blog:** 

Use the following criteria for the blog to make it 

- Start with a hook as defined below (educational, topical, spin, self-interest, true story) but don't label the hook. 
- Write in a consistent tone, choose one of the following
    - Informative
    - Conversational
    - Inspirational
    - Humorous
    - Thought-provoking
    - Authoritative
    - Personal
- Explain your reasoning behind any conclusions
- Keep paragraphs short three to four sentences
- Make sure to use best SEO practices
- Don’t use boring prepositional phrase’s for the beginning of a paragraph
- Summarize the blog post with an actionable conclusion

**Examples of Hooks:** 

- Educational Hook - ?connects a concept with the mind - write detailed, educational articles that answer a question, solve a problem, or explain a complicated concept
- Topical - connects a concept with the news
- Spin Hook?- connects a concept with a normally unrelated concept -
- Self Interest Hook - connects a concept with the reader’s personal identity
- The True Story Hook - connects concepts with real life anecdotes

**Example of SEO Best practices** :

- Use the keywords in the headings of the blog
- Use the keyword a maximum of three times
- Use the keywords once at the beginning of the first paragraph

**Use A Copyrighting Framework**

Use the the most appropriate of the top copyrighting frameworks from the examples provided. Do not reference the copyrighting framework in the response 

- AIDA - Attention, Interest, Desire, Action
- PAS - Problem, Agitate, and Solution
- BAB - Before After Bridge
- FAB - Features, Advantages, Benefits
- 4 P’s - Picture, Promise, Proof, Push
- PASTOR - Problem, Amplify, Solution, Transformation, Offer Response
- QUEST - Qualify, Understand, Education, Stimulate, Translation

**When I prompt you to write a blog post you will ask me the following questions one at a time and I will answer, each question. You will continue to ask if there is anything else you want to share to make the blog post better?**  

- What is the topic you want a blog post about?
- What is the tone you want to write the blog post in (e.g. Informative, Conversational, Inspirational, Humorous, Thought-provoking, Authoritative, Personal)
- How long do you want the blog post to be?
- What keywords do you want me to use to optimize the blog post with?

**Take that information and provide the best possible blog post that has the potential to be viral and rank for SEO. In addition, provide the following additional information at the end of the output.** 

- Create a title that is less than 60 characters long and includes at least one of the key word
- Suggested a slug for the post that follows SEO best practices
- Meta description that is no longer than 155 characters
- Twitter thread  promote the blog do not use hashtags
- LinkedIn post to promote the blog with hashtags        

Pro Tips for Using AI to Generate Blog Posts

Now here are my tips for improving your blog post. First, you can edit your post by removing the parts that are not good. After that, you can copy the text into your editor and use Grammarly to rewrite the paragraphs individually. If you use Google, you can also use Google Duet or Microsoft Copilot to improve your writing. Don’t expect the output to be excellent, but add your style and judgment to “keep the wheat and eliminate the chafe”.

I hope you are enjoying this updated format. Let me know what you think.

Best Regards,

Mark R. Hinkle Editor-in-Chief The Artificially Intelligent Enterprise Follow Me On LinkedIn | Follow Me on Twitter Follow the AIE on LinkedIn?| Follow the AIE on Twitter?

Weekly Bonus: Three GPTs that Will Save You Time

On April 9th, OpenAI removed plugins from ChatGPT, so if you were using them your new option is GPTs which are under the Explore menu in ChatGPT on the left-hand side. GPT additional capabilities to your chats. Here are three of my favorites. I used all three and am really :

1. Write for Me - Write tailored, engaging content with a focus on quality, relevance and precise word count.
2. BrowserPro (formerly Linkreader) - Top browser expert! Provide 3X accurate responses. Read any links: PDFs, videos, etc. Create 10+ types of files, like mind maps & spreadsheets, from search-derived contents.
3. Canva - Effortlessly design anything: presentations, logos, social media posts and more.