Up close with ChatGPT
Interaction with the chatbot shows the power of generative artificial intelligence
ChatGPT has become a global sensation since its launch in November 2022. Unlike previous generations of chatbots which were restricted to pre-programmed responses, ChatGPT is powered by generative artificial intelligence that uses large language models to generate comprehensive and nuanced responses to queries in a manner similar to human conversation.
More than one million users registered with its online website in the first five days after ChatGPT’s launch, and over 100 million people globally became active users within the first two months, according to an analysis by Swiss bank UBS. It has broken all records as the fastest-growing app in history.
After reading the overwhelming reviews in the media and hearing glowing feedback from friends and family members, I was keen to give it a try.
But there was a small hiccup in getting started. OpenAI, the creator of ChatGPT, doesn’t allow access to ChatGPT from Hong Kong, where I am based. Countries on the barred list include China, Iran, North Korea and Russia. This meant I had to access the app by connecting to a foreign network via a VPN and registering with OpenAI with an overseas telephone number for authentication. That delayed my debut on the app.
This issue has since been resolved, when in March after a new app called Poe was made available to Hong Kong residents that allows access to ChatGPT, as well as Sage, Claude and several other AI chatbots.
As a first-time user, I was immediately wowed by ChatGPT’s ability and speed in generating high quality responses to my queries covering a wide range of topics.
My wife and I were planning a holiday in Germany so I asked ChatGPT to recommend five restaurants within a 20 minute drive from the Grand Hyatt Hotel Berlin where we would be staying.
Within seconds, I received suggestions for three Michelin starred restaurants serving different cuisines – one German, one contemporary European and the third Asian-inspired – as well as two other trendy French and Dutch restaurants. Based on their websites, they would all be very good options, serving some interesting dishes.
ChatGPT also prompted me on other questions that might be helpful, such as more casual eateries, getting around Berlin, and so on.
Of God, and a job recommendation
After getting satisfactory answers to a couple more questions, it was time for a profound one: “Who is God?”
ChatGPT explained that the concept of God varies across different religions and belief systems, highlighting the differences in interpretation by monotheistic religions such as Christianity, Islam and Judaism, polytheistic religions such as Hinduism, and spiritual enlightenment systems such as Buddhism.
“Ultimately, the concept of God is deeply personal and varies widely depending on individual beliefs and cultural traditions,” it concluded. Well said!
Then I changed tack and asked it to provide some pointers to write a job recommendation letter for a female ex-colleague.
Again, within seconds came four suggestions, along with elaboration: first, start with a short statement summarising her strengths, then share specific examples and highlight her unique qualities, point out the impact she made on the company, and finish off with an enthusiastic endorsement of her.
These were sound suggestions. I drafted a short recommendation based on the pointers and asked ChatGPT to review it. The revised version from the chatbot was well-organised with good word choices and nicely written sentences. It was definitely an improvement on mine.
But amazement aside, I felt that the revision was a bit like an extract from some commercial templates. It wasn’t my writing style. So, rather than accepting it as the final product, I picked bits and pieces that I liked and used them to revise my original draft.
I gave my feedback to ChatGPT and asked it to do another revision. This process was repeated for another round until a final version was produced that I was happy with.
My ex-colleague was thrilled with my recommendation letter. Suffice to say I didn’t tell her it was produced in collaboration with an AI-powered chatbot!
Notable issues
The power of generative artificial intelligence is evident through these brief experiences with ChatGPT.
For starters, it’s very easy to use. It interacts with you in conversational text as if you are talking with another person. You know it’s not human, but it feels like you’re conversing with one.
In fact, it’s like you’re talking to someone who is incredibly knowledgeable and quick-witted. The AI chatbot can answer a whole range of questions from the simple to the sophisticated in a matter of seconds and assist you with all sorts of tasks.
Apart from using ChatGPT for research, many users are excited about its intelligence in replying to emails, summarising documents, writing essays, brainstorming ideas, creating itineraries, composing music or coding programmes.
The chatbot also puts a personal touch to its responses. For example, after suggesting five restaurants to try on my vacation, it finished with a personal note: “I hope these suggestions are helpful and wish you and your wife a wonderful trip to Berlin.”
It is a sophisticated tool. The more specific and descriptive the context you provide, the more precise and customised the output it will produce. You tell it what you think of the generated response, and it will then improve the output based on your feedback. The process is an interactive collaboration between you and the AI tool.
However, ChatGPT has some notable issues. It’s trained on millions of pieces of data and information from various sources to produce responses. As a result, whether it’s ChatGPT, Sage or another AI chatbot, they all generate quite similar content if given the same prompts, probably because they are trained on similar information. Humans, on the other hand, are likely to generate more diverse ideas even when given similar experiences.
A related issue is potential bias. These AI chatbots can generate biased content depending on the nature of the training information they were fed on. This may potentially perpetuate harmful stereotypes and biases. OpenAI acknowledges the limitations and warns that ChatGPT “may make incorrect statements” and “does not have knowledge of events after 2021”.
And then too, the interactions are so human-like that users may not be aware they are communicating with a machine, which leads to the ethical concern of potential deception if the content were presented as being created by a human.
Lastly, although ChatGPT dramatically reduces the time spent researching a subject, the information obtained may not be accurate. By contrast, Google search is a time-consuming process that requires ploughing through a sea of information with duplicate or conflicting views from different sources, but it will allow you to compare, verify, and fact check the accuracy.
Huge potential
Despite these issues, I can see enormous potential with ChatGPT. What I have experienced thus far is only a glimpse of the future of artificial intelligence. I’m sure over time the shortcomings will be minimised and its strengths enhanced.
I asked ChatGPT how it may be applied to the financial industry.
It suggested four examples. First was in customer service, to help banks and financial institutions reduce the workload of their customer service representatives and provide faster and more efficient service to their clients. The second was in personal finance management for individuals. The third was in fraud prevention, to help banks and financial institutions detect fraudulent transactions early and prevent losses, and lastly in trading and investment, to help traders and investors make informed decisions about their investments.
“Overall, AI chatbots have the potential to improve customer service, increase efficiency, and reduce costs in the financial industry,” ChatGPT said.
What about examples of potential drawbacks of using the AI chatbot in the financial industry?
Firstly, security concerns require that chatbots be designed with strong security measures to protect against data breaches, identity theft, and other security threats. Then, limited personal touch means some customers may prefer to speak to a human customer service representative, particularly when dealing with sensitive financial matters or complex issues.
Moreover, chatbots may not be able to fully understand the context of a user’s question or provide personalised advice based on the user’s specific financial situation. And finally, chatbots require a reliable internet connection and may be subject to technical glitches or downtime, impacting the customer experience.
“Financial institutions can balance the use of chatbots and human customer service, providing customers with a seamless and personalised experience that meets their needs and expectations,” ChatGPT concluded.
Generative AI is kickstarting a new industrial revolution. It’s time for different industries to implement strategies to leverage the potential while mitigating the risks.
*This article was published in Asia Asset Management’s July 2023 magazine under the same title.