ChatGPT is, in a manner of speaking, the belle of the AI ball. Why’s that? Well sure, there’s the allure, but also (surprisingly, given its recent debut) its all-but universality. This novelty is super tempting for casual users and corporations alike, but people are also dialing back. Seeds of concern have been planted about abuses of ChatGPT; as in, its potential to perpetuate harmful biases and misinformation, as well as the issues of ethics and privacy violations.
At the new Generative AI Expo, taking place at the Broward County Convention Center in Fort Lauderdale, FL from February 14-16, the kick-off discussion covered these points in detail; the shininess of ChatGPT, pros and cons, and whether or not it should be seen as the future of work.
Helming the discussion were Jeff Dworkin, Managing Director and Chairman of Cloud Voice Alliance, and David Jodoin, CTO of NYNJA Technologies. It was moderated by Jon Arnold, Tech Analyst and Principal of J Arnold & Associates.
Here, for your long-story-short needs, is the gist of the discussion.
After Arnold’s introductions, Jodoin opened with an audience show-of-hands. Who was present to learn about ChatGPT for the first time? Who was concerned about how ChatGPT would affect their organization, and whether or not they should get involved? And, moreover, who knew ChatGPT trained and what are the impacts therein?
In the room, several hands were raised by ChatGPT first-timers. More hands shot up, however – a couple dozen more – with attendee concerns about ChatGPT, training AI in general, and the relevancy of human involvement.
Jodoin and Dworkin had answers.
In short, ChatGPT is just one implementation that the OpenAI model (i.e. algorithms and engines) powers. With ChatGPT, writers can use it as a search engine for quick research, students can use it for essays, and organizations can utilize it to help with their customer bases.
But the common theme of today’s discussion?
Validate. Always validate.
ChatGPT can indeed skillfully converse, write content and code, answer inquiries, and more ad infinitum. It talks like a human and generates confident responses.
But it’s not infallible, and its domain is not universal.
Per today’s speakers, ChatGPT was trained using neural network architecture, and the training datasets consist of a large corpus of preprocessed text from the internet, as to omit sensitive or personally identifying info. It knows a heck of a lot, but not everything it renders is gospel. It’s not all-encompassing.
So, circling back, writers using it as a search engine should verify the facts. Students shouldn’t use it to complete the essay-writing for them; they should double-check, add to, and modify before submitting the end result to their educators. Organizations utilizing it for their customer bases should keep, at the very forefront, brand voice and customer expectations. (Because ChatGPT can’t replicate them.)
As Jodoin noted, “You can certainly take OpenAI model and train it to your specific needs, but it’s still – from an AI-in-progress standpoint – essential to set parameters and authenticate what’s generated.”
“ChatGPT lifts a lot for us,” Dworkin said, “but it’s not an end-all, be-all replacement of the lifters. There are nuances that it isn’t aware of. The important parts of our jobs still come down to our brains, which is why ChatGPT won’t replace us. So, when using it, approach with a critical eye rather than oversized rose-colored glasses.”
More about Generative AI Expo can be found here.
Edited by Alex Passett