How to extract business value from ChatGPT
How much trust should your business be putting in ChatGPT? Horatiu Petrescu says don’t focus on the tool, focus on who is using it. The question of exactly how much […]
How much trust should your business be putting in ChatGPT? Horatiu Petrescu says don’t focus on the tool, focus on who is using it.
The question of exactly how much trust we should put into ChatGPT has a simple answer, and a more nuanced one.
For starters, it’s necessary to understand what ChatGPT is.
At the most basic level, ChatGPT is a tool. Ask yourself how much you might trust any tool, say a spade, and you’ll have the simple answer to how much you should trust ChatGPT.
The point is that tools are neither trustworthy nor untrustworthy. Instead, a tool is only as trustworthy as the person who holds it. How a person wields a spade matters because the tool can be used for digging a hole and it can be used as a weapon.
This is where the more nuanced answer surrounding trust and ChatGPT starts emerging.
We’ve already seen examples of how in the workplace ChatGPT’s shortcomings have seen its users dig themselves into deep holes. Take the lawyer who relied on ChatGPT for help; it turned out the ‘case law references’ conjured up by ChatGPT simply didn’t exist.
While ChatGPT’s output is superficially impressive, particularly to the layman, our ignorance is its biggest advantage. The best test of its abilities is asking about something on which you are already an expert. You may find that the answers can’t match that of a human with deep knowledge on a topic – an issue if depth and breadth are the essence of your needs.
The more you use it, the more you’ll notice chinks emerging in its armour. Beyond simply making things up, ChatGPT provides a formulaic output, which while initially highly plausible, quickly falls into a standard and dispassionate format.
Subtlety, humour, references to current affairs, the use of regional or local dialects or mentions of customs and culture are often nowhere to be found. This is the stuff that makes writing sincere and engaging as opposed to technically good. While these elements may be difficult to detect, they can make all the difference when crafting persuasive communications – whether that be for business, or domains like journalism, public relations and professional writing.
And then we get to mathematics, where there are multiple examples emerging showing ChatGPT just isn’t very good at this endeavour.
There’s good reason for these shortcomings. As a generative Large Language Model, ChatGPT is exceedingly capable when it comes to grammar, syntax, and other technical features of language. It draws on existing public bodies of knowledge. It knows which words are statistically most likely to follow one another. It knows how we express ourselves and seeks to mimic it.
What these experiments show is that like any good tool, it’s advised to ‘read the manual’ before putting it to use. Most of us won’t have too much trouble using a spade and figuring out the business end probably won’t require instruction. ChatGPT, on the other hand, is highly sophisticated. Just like you’d treat a power tool with more respect, the real question quickly becomes not ‘how much should I trust it’, but ‘how careful should I be when using it’.
As a general rule, if you feel you should conceal your use of ChatGPT for any reason, then you probably should not be using it at work.
Extracting value
ChatGPT, however, does have some use in the workplace. If you are aware of the technology’s limitations and weaknesses, you’re in a position to start getting value out of it.
For example, it’s exceptionally good at generating trivial content which for us humans might be time-consuming: it will deliver structures for letters, PowerPoint presentations, annual reports, and so on, fast and with ease.
It is highly useful in summarising your existing work, generating sub-headings or helping a writer address their block by rephrasing or suggesting new directions.
Like a spade, or really any tool, it can be immensely helpful with some tasks, and completely useless for others like mathematical equations or predicting results in the future.
Bear in mind that new tools routinely enter the classroom and the workplace. Back in the 1980s and 1990s, a wave of worry accompanied the calculator. We feared the rise of the personal computer, too. But just as it is with ChatGPT, these tools made us more efficient and effective once we knew how to use them to our advantage.
And in that respect, nothing has changed.
Horatiu Petrescu (pictured) is security consultant at Aura Information Security.