With AI-powered ChatGPT, the watchword is verify, don't blindly trust

If you’re stumped on an idea, need a quick blog post or even need a brochure translated into another language, you could consider trying the popular conversational artificial intelligence-powered chatbot ChatGPT.

It may be the fastest growing consumer application in history, but while it has many uses, from creating content to generating ideas, it has certain limitations, experts say.

“AI could have ‘hallucinations’ and make things up,” says Paul Trapani, president of Long Island Software and Technology Network (LISTnet), a tech advocacy group and co-founder of PassTech Development, a software development and consulting company. “It’s not always accurate and at times it has added information I didn’t add.” 

He uses ChatGPT, launched by OpenAI in November, for numerous purposes including creating event listings for LISTnet gatherings.

On a recent listing, it added on its own that an event would have a DJ, which Trapani never  input.

“I would say for now humans should still be involved in at least vetting the results,” Trapani said.

OpenAI has forecast improved accuracy with its new release of GPT-4, an update to the technology powering ChatGPT. In a post, the firm said “GPT-4 is more reliable, creative and able to handle much more nuanced instructions than GPT-3.5.”

Still, OpenAI cautioned “it still is not fully reliable" — it “hallucinates” facts and makes errors in reasoning.

GPT-4 will initially be available to ChatGPT Plus subscribers who pay $20 a month, but not yet available to those using the free version of ChatGPT.

ChatGPT brainstorming

Trapani uses the paid version. He’s tasked ChatGPT to brainstorm ideas for events, create listings and even to analyze programming code.

Still, he always double-checks results and advises others to do the same.

The free version works relatively well, says Derek Peterson, CEO of Ronkonkoma-based Soter Technologies, a software and hardware developer of  high-tech vaping and bullying detection systems.

“You might have to edit the results five percent, but overall it’s pretty accurate,” he says.

It does have its shortcomings, and he's found it's not well-suited for technical writing.

His marketing and sales team uses it to write content for their website, blogs and newsletters and for crafting sales proposals with bullet points.

He can also take some of his current brochures and ChatGPT will translate it into other languages.

“In business you want to maximize your productivity and that’s what it’s able to do for you,” Peterson says.

“What ChatGPT allows us to do is instead of being writers, we become editors,” says Constantine Gus Spathis, founder/CEO of Xogito Group, a Huntington-based product development company.

Like others, he’s seen it make errors. But one interesting element is you can tell it to adopt a certain writing style, such as to create copy that’s either a bit a more cheeky or serious in tone, he says.

He uses it often for summarization of meetings, specifically asking it to break out important action items from speech-to-text software-generated transcripts of meetings.

Its capabilities are broad, and the underlying technology behind ChatGPT “is going to radically change business operations,” Spathis says.

His own company is using that underlying technology — GPT-3 and soon GPT-4 — to build AI-enabled products for different uses for clients from industrial product companies to marketing companies.

Austin Williams, a Hauppauge-based advertising and digital marketing agency, has also used the underlying GPT technology to build a live chat tool that will be available to clients by mid-April, says chief innovation officer Andrew Catalano.

He said they developed it because many clients were frustrated by the off-the-shelf live chat tools currently out there and felt they were limited and not very humanlike.

Separately, Austin Williams uses ChatGPT in its office for inspiration or to draft content that’s search engine-optimized.

Catalano has concerns that down the line, search engines will start to recognize content created by AI technology like ChatGPT and start to block it since search engines always push for original content.

That hasn’t happened yet and so far it’s been a good tool for them, Catalano says.

But he does agree that “it’s not entirely accurate and you always need to fact-check the output for the most part.”

Alana D’Elia, a Westbury-based digital marketer, is always cautious when using it.

“It’s kind of a first-draft brainstorming tool,” she says.

When she used it to create a blog for a personal injury attorney’s website, “it quoted all the wrong law.”

“A subject-matter expert really has to review it,” D’Elia says. But, on the plus side, it hasn’t created plagiarized content. She’s always super cautious about that and uses Grammarly’s plagiarism checker.

She’s asked ChatGPT to come up with taglines, Google Ads headings, website content, blogs, and advertising and branding copy.

"It’s only as good as your query,” D’Elia says. “There’s an art to querying that people have to learn.”

By Jamie Herzlich

jherzlich@aol.com

Herzlich writes the Small Business column in Newsday.