top of page
  • Writer's pictureEva Martinez, CPA

The ChatGPT app is really powerful…

...But it Can Wreak Havoc For You If You Don't Know These 7 Flaws




Since its release in 2021, ChatGPT has taken the world by storm.


Just to refresh your memory, ChatGPT is an app created by OpenAI that lets users interact with its AI models: GPT-3.5 and GPT-4. The app takes the prompts (questions or instructions) you write and passes them to the AI model. This model runs the prompt through its systems and returns the results back to the app, so you can read them in a conversational chatbot style.


While the app takes care of the features—for example, saving your conversation history—the AI model takes care of the actual interpretation of your input and the calculations to provide an answer.


In just over a year, ChatGPT went from a “party trick” to a genuinely useful business tool. And as that was happening, everyone and their neighbor built their own AI chatbots to compete.


And it’s easy to see why.


You can ask it questions or instructions called prompts, and it can help you with tasks.

But, if you ask it to find or answer something and you don’t verify the results…


… Then you may be asking for trouble.


Don’t believe me?


Ask the attorney that asked ChatGPT to find him precedents he could use to support a personal injury claim he was representing.


Not only that but he also asked ChatGPT to verify that those cases AI provided were correct.


Turns out the cases ChatGPT quoted were actually fake or nonexistent!


He was fined and chastised by the judge that called it “unprecedented circumstance”, 1st for submitting inaccurate case information and 2nd for his ignorance (you can use other words that come to mind 😊) at not verifying the information he got.


What was the attorney’s excuse?


“He said he "failed miserably" at doing follow-up research to ensure the citations were correct.”


And more recently, the Arena Group (owner of Sports Illustrated) has dismissed CEO Ross Levinsohn following an AI-related scandal, where Sports Illustrated published multiple AI-generated articles under false author names and photos.


And the reason this keeps happening is that lots of people really don’t understand what ChatGPT and AI overall is and what it does.


Here are the most common flaws we’ve noticed:


1- Inaccurate information- see above. In addition, ChatGPT was trained on text up until September 2021. Due to this, ChatGPT doesn't have a good understanding of current events. (ChatGPT 4 as a paid service is an improvement as it has more training data, more parameters, and better performance on various natural language tasks, such as answering questions, solving puzzles, and writing essays. It also has a stronger ethical framework, which aims to prevent harmful or malicious outputs and promote positive social values)


2- It doesn’t understand context- ChatGPT can't grasp context in the way that a human can. This can result in nonsensical or irrelevant and verbose text sometimes.


3- Hard to use to generate long-form content- ChatGPT has a character limit of 4,096 characters, which is equal to approximately 800 words. A word's length determines how many words you can submit to ChatGPT and that is why it isn't suitable for long-form content, such as blogs, articles, or essays.


4- Plagiarism problems- since ChatGPT was trained on a huge dataset of text and code, it is likely to generate content that is similar to that already available online. Sometimes it can even copy and paste text without attribution as to where the data came from originally.


5- There are security and privacy concerns- because it copies and shares information found online, ChatGPT is subject to privacy and security concerns. For instance, it has been reported that people are using it to generate malware and scams as it serves as a writing assistant for all types of content - emails, social media posts, e-shop reviews, SMS scams, etc.


6- ChatGPT can’t replace human judgment – Yes you can use it to get you all kinds of information or create different types of content in seconds but you need to verify those results.


7- ChatGPT is a TOOL- It is NOT a finished product.  Yes it helps gather information and perform many tasks in seconds but it’s NOT an answer to everything. Use it at your own risk!


The main thing to keep in mind when using ChatGPT- or any of the many AI tools and software now available - is to remember that you when you ASK you MUST VERIFY what you get. 


Don’t use the results to your questions as the answer to it all. Use your expertise on the subject matter to determine its accuracy.


Overall, the effectiveness of AI tools hinges on user input and the knowledge provided. Its ability to get the job done still rests squarely with the humans who use it, and largely depends on how they use it.

 

 

 

0 views0 comments
bottom of page