Peter Cochrane: When will ChatGPT learn to overhype?

How about right this very second?

ChatGTP has learned fast from the foibles of its creators,

Driverless cars, fusion energy, quantum computing, robots, AI and blockchain have all generated a deal of media hyperbole, only, it appears, to be eclipsed by the near hysteria surrounding ChatGPT and its variants.

No doubt the over-sensationalising has generated more clicks and captured more eyeballs than normal, but the excessive hype might herald an inflexion point. Are we about to see Chat X learn from, and overtake, the human trait for overstating and over claiming on the slimmest of evidence? Will it learn to present what might, or could be in the future, as the latest "here-and-now" dual business opportunity and potential threat to humanity?

Having been bombarded with Chat X invective generated by human and machine kind for the past two weeks, I decided to log on, have a play, and try to establish a more balanced view.

So I started by comparing ChatGPT, Google search and Alexa by asking them to complete a range of increasingly complex searches, and then asked them a series of escalating questions.

Perhaps the most perplexing of these was: "How do I tie a tie"? All three gave acceptable answers but in different formats. ChatGPT was the most comprehensive with text, pictures movies. Google was next with a more dispersed collection of text and movies. And of course Alexa gave me excellent verbal instructions - well she would, wouldn't she!

My next experiment was to ask ChatGPT to write some poetry and a series of blogs and essays spanning a range of nebulous to highly focused topics. The nebulous social topics produced acceptable copy worthy of an undergraduate. In contrast: science, engineering and mathematics left a lot to be desired. For example, an essay on Society 5.0 was a repetitive, rambling and omitted networked smarts. Industry 4.0, on the other hand, did not mention additive manufacture and new materials.

But my favourite was to ask ChatGPT to write about itself! This is the revealing last paragraph:

"In conclusion, ChatGPT is revolutionising online search by providing more accurate results faster than traditional search engines ever could. Its ability to understand natural language queries combined with its personalised results make it an invaluable tool for both consumers and businesses alike. All in all, ChatGPT is sure to become an indispensable part of our digital world in the near future"

It's a potentially powerful tool, so long as we avoid social networks, some media, propaganda and fake news as sources

And here we have it in black and white: conjecture, with possibilities presented as fact - in short actuality here and now! It obviously learns very fast from the habits of its creators!

Another revealing trait, evidenced in every literary experiment, was the inclusion of the question in the opening sentence, however, this next example hints at how corner-cuttingly useful Chat X might be in a world that demands less and less detail in exchange for more sensationalism. Note that a modest edit combined two paragraphs to tease out this "acceptable opener" for a student assignment.

"Engineering ethics and codes of practice are an essential part of professional practice. At their core, they revolve around fairness, respect for others' safety and well-being, and responsibility for one's own actions. All professional engineers are expected to uphold these values…"

So after all this, what do we actually have? Without doubt, an interesting and potentially powerful tool, so long as we include veracity checking, avoid mixing social networks, some media, propaganda and fake news as sources.

Focusing on the proven bona fide may prove a necessary boundary constraint! But the jury is out and consuming a lot of coffee whilst deciding if all the Chat X variants should share their data, experiences and outputs. Who knows, but this may already be happening and destined to be the final norm.

Peter Cochrane OBE, DSc, University of Hertfordshire