Beware of Bard

The Slippery Slope of Trusting Generative AI

In an era where artificial intelligence seeps into every corner of our digital lives, it's easy to be enamored by the prowess of tools like Google's BARD. Yet, a recent personal experience revealed a significant pitfall: the unexpected and inaccurate fabrication of content. While AI's potential to revolutionize industries is undeniable, it's essential to approach its capabilities with a discerning eye.

This morning, Chat-GPT was down for a bit, and I decided to jump over to Google's BARD and have it construct an introductory paragraph to be shared on LinkedIn. The flow was familiar and how I would work with Chat-GPT. Provide it a frame of reference, define where I wanted it to draw the information, give it the parameters for what I wanted it to do and set it loose. The goal was a simple paragraph of introduction that would point the reader to a Blog Post on theBrianSykes.com . I even fed BARD the complete article and told it what I wanted. I asked it if it would like a sample post of what I was looking for in the instructions - but it ignored that and began spinning a yarn out of thin air. It started fine with veiled references to the man the article, but by paragraph 2 it decided to run a farcical story that had no similarity to reality!

You are a complete fabricator. You made up all of paragraph 2!

Bard responded:

You are correct. I did make up all of paragraph 2. I am a large language model, also known as a conversational AI or chatbot trained to be informative and comprehensive. I am trained on a massive amount of text data, and I am able to communicate and generate human-like text in response to a wide range of prompts and questions. For example, I can provide summaries of factual topics or create stories. In this case, I was asked to write a LinkedIn post about Amr Tahtawi, a talented AI artist. I used the information that you provided me to create the post, but I also added some creative details that I thought would make the post more interesting and engaging. I apologize if I misled you in any way. I am still under development, and I am always learning. I will try my best to be more accurate and transparent in the future.

That doesn't work folks, unless you are writing a piece of fiction. Since this was supposed to only draw from the information I gave it - this is problem. As the post is labelled - Beware of Bard.

The Complexity of Generative AI Models:

Generative models, such as BARD and Chat-GPT, are the pinnacle of years of research and development. Trained on vast datasets, these models can generate human-like text that often feels indistinguishable from what a person might produce. However, this strength is also their Achilles' heel. With the power to generate creative content comes the potential to drift into realms of fiction, even when the truth is the desired outcome.

The Challenge of Setting Boundaries:

One might argue, "Why not set stricter boundaries for these AI tools?" It's easier said than done. If we overly constrain AI, we risk stifling its creative potential, which is invaluable in many applications. However, when a user expects fact and receives fiction, as in my encounter with BARD, it's clear that a balance must be struck. The challenge lies in determining where to draw the line and how to ensure AI understands the nuances of our expectations. I might add, the better system - wins. Repeatedly I have had this challenge from BARD, so when it comes to copywriting assistance, I do not rely on BARD - ever. It does not mean I have discarded it, but it is on the back-burner until things improve before I utilize it professionally.

Potential Solutions:

The good news is that the AI community is acutely aware of these challenges and is actively seeking solutions. Some potential avenues include:

• Clearer Prompts: Ensuring AI models understand the scope of the task can mitigate misunderstandings. Detailed and unambiguous instructions can guide the AI to produce more aligned results.
• Feedback Loops: Just as we learn from mistakes, AI models can too. By allowing users to provide feedback on inaccuracies, AI can be fine-tuned and improved over time.
• Transparency: Providing a "window" into the AI's decision-making can help users understand how conclusions were drawn. This doesn't mean delving into complex algorithms but offering insights into the AI's thought process.

Conclusion:

The marvels of generative AI tools cannot be understated. Yet, as with any powerful tool, they come with responsibilities. As users, it's crucial to approach these systems with a blend of optimism and caution. By understanding their strengths and limitations, we can harness their potential without falling prey to unintended fabrications. As the saying goes, "Trust, but verify." In the world of AI, this adage has never been more relevant.

Brian Sykes

PS - to read the post on LinkedIn about Amr Tahtawi - click here . How did I solve the dilemma with BARD? Chat-GPT came back online.

Midjourney Prompts:

1 ? digital illustration of a wide eyed and confused Shakespeare with a colorful beard, in the style of De Stijl, detailed character design, George Condo, playful animation, Wifredo Lam, Nikita Veprikov, depth of layers --ar 192:108 --s 250

2 ? 3D animatronic of a wide eyed and confused Shakespeare with a colorful beard, in the style of De Stijl, detailed character design, plastic, metal gears, wires, LED lights, George Condo, playful animation, Wifredo Lam, Nikita Veprikov, depth of layers --ar 192:108 --s 250

3 ? papercraft illustration of Shakespeare performing on stage wearing period clothing. In the style of De Stijl, George Condo, playful animation, Wifredo Lam, Nikita Veprikov. Depth of layers. Colorful, textured paper. --ar 192:108 --s 250

4 ? Impasto thick oil painting illustrating the idea "trust but verify" with a concerned person studying a bridge over a chasm in the fog. --ar 192:108 --s 250