ChatGPT providess answers to all sorts of questions based on being able to string the most likely words together found on the Internet. While it may be useful for writing poems, job descriptions and novels, it has been terrible when it comes to researching facts. In my experience it’s been so wrong, it may be less of a threat than it seems. After all, if a source keeps getting things wrong, its credibility is questioned and that source is eventually dismissed or ignored.
If Google Maps kept directing you to the wrong location you’d stop use it, even if only did it 10% of the time. If you’re going to rely on something as important as ChatGPT, it needs to be right nearly all of the time.
In a recent court case, as described in the NY TImes, a lawyer cited a series of previous court cases to bolster his arguement during a injury suit brought by his client against an airline for bumping into his knee with a food cart. When the airline’s lawyer said he could not find those cases, the judge asked the suing lawyer for more specifics. It was discovered the cases sited didn’t exist; they were all made up by ChatGPT. Even the lawyer that used ChatGPT was surprised.
That’s been my experience as well using ChatGPT. When I first asked it for my bio it said I wrote technology articles appearing in the New York Times and The Economist. ChatGPT was good at making up facts with such a high level certainty and specificity, we all believed they must be true.
Recently, while doing research on high resolution streaming music services, ChatGPT explained how they dealt with adverse network conditions. The explanation was totally made up, bearing no relation to the facts. That’s because ChatGPT has no idea of what’s real and what’s imagined, and no ability to distinguish between the two. It’s basically just guessing.
Wouldn’t yout think that the more people use this product, the more will learn how unreliable it is and no longer use it?
The real question is whether ChatGPT can improve itself enough to not just guess, or whether its basic design is inherently flawed because of the methodology it uses. I’m not an expert here, but once its aura wears off, maybe we’ll dismiss it as being a novelty and not the technology breakthrough being attributed to it, at least with regard to its accuracy and usefulness.