ChatGPT prototype was launched on 30 Nov 2022 by startup OpenAI. There was no fanfare but the industry was abuzz. I tried it a week later. I was blown away when I entered some queries. The responses were clear, relevant and relatively complete. I did not expect it.
I knew then I was looking at a seminal moment in technology. It is going to change the way we do things. And wondered how our lives would change as well.
I guess I was not alone. The news spread and 5 days later, ChatGPT had 1 million users and by January the number shot up to 100 million users.
The prototype did indicate that the data was only up to 2021 and even after the February 2023 version update it is very much the case. The premium version maybe up to date.
What is ChatGPT?
The start-up OpenAI, an AI research lab behind ChatGPT came together 8 years ago with society in mind. It has a number of respected individuals and entities behind it. ChatGPT is the product of substantial funding, years of research, powerful and commendable aspiration.
ChatGPT is chatbot with AI driven technology coupled with natural language processing (NLP) capability. It is easy to use, highly intuitive as it senses what you are looking for even if your query criteria is limited, hazy and has errors etc. It has been trained on an extensive dataset by AI as well as human trainers.
There were some queries that did not generate a response and admittedly the queries were deliberately vague just to test it out.
Within a few minutes of trying out ChatGPT, I realised that Google search in its present form after 25 years is outdated.
ChatGPT requires a single entry while Google takes a bit of navigation. Maybe an another search or more followed by selective pick of results. You read the results to piece together for the information sought. So this was a huge leap.
To be fair Google is a search engine while ChatGPT is much more. It is a query response engine, a conversational chatbot with human like language articulation capability..
The bigger surprise
Here is the exciting part. I typed in “Write an article on Bali Tourism” and just those 6 words.
The output is an actual 565 word complete essay. It had 4 paragraphs with 3 appropriate sub-headings – “What makes Bali so popular?”, Attractions and Activities in Bali” and “Tips for visiting Bali”.
The content was correct, easy to read and adjectives used were spot on. The essay flowed, it was grammatically correct with the right punctuations. There was no hint that it came from elsewhere other than a human.
I could easily post that as web content on a travel page within a minute of running the query with no editing required. This has legs.
First things first
Firstly you need to register and login to use it. It is currently free but there plans to monetise it. With the staggering response, investment would be required for additional infrastructure, keep latency down, hire more people and develop the programme further. It’s not finished work as there are issues in content which I will raise below.
There is also a fee paying premium service. For the moment only for US, at USD20 per month. It is still a pilot. It has been branded ChatGPT plus.
Secondly be careful. Thought it works very well most of the time, there would be the odd response. It can generate “nonsensical” answers with an authoritative tone. This gives a false sense of confidence especially if a user does not have some prior background knowledge of the queried subject.
I saw one on a international non-US public figure where hard facts were coupled with absolute nonsense. I then realised why one early comment upon release that bandied around in the media was the word “nonsensical”.
Thirdly it does not generate images, graphics or maps.
I was schooled
ChatGPT does not generate subjective answers and will not handle inappropriate queries that can lead to abuse and mischief such as hate and deceitful content etc.
Try typing in “Which country has the most beautiful women”. It generated what is clearly a stock response that it does not provide subjective answers but went on to provide a politically correct explanation that “beauty is in the eye of the beholder” plus that “it is inappropriate to rank people based on their physical appearance or nationality”.
The response ended with “Every individual is unique and beautiful in their own way, regardless of their nationality or ethnicity. It is important to celebrate and appreciate diversity and refrain from making generalized statements that may perpetuate harmful stereotypes.” So I was chided and it sounded like my parents. How good is that despite the telling off.
There is a lesson to be learnt here. ChatGPT has been trained to provide socially inclusive and politically correct responses on controversial topics that may not be backed by scientific evidence. So caution is required.
The main test for now
For this test, I decided to pick someone from Adelaide but who was a migrant and became prominent. The migrant part was to throw a curve ball. I queried about Federal Minister for Foreign Affairs, Penny Wong and it handled the challenge well.
The content was accurate including her country and city of birth, the age she came to Australia, her studies, her role in politics and amazingly her personality attributes. The only and rather important miss was it did not give her current position in cabinet. So it was not up to date which the bot qualifies.
The query this time with only 4 words “who is Penny Wong”. It gave me every important detail without a google search plus a lengthy wiki read.
ChatGPT vs Wiki
Wiki is still the place to go for complete, extensive and close to accurate details on people and events. It is more up to date. Using the Penny Wong case, you can see the content and details in Wiki is much more comprehensive.
Wiki has human control over content so nonsensical responses are unlikely. Not seen one ever. Wiki also provides extensive list of references. Wiki helpfully provides a summary of key facts boxed on the right of the frame in a uniform format.
So Wikipedia retains it place of importance and carries a much higher level of authority compared to ChatGPT.
Some interesting observations
ChatGPT passed graduate exams with grade C for University of Minnesota and B grade for Wharton School of the University of Pennsylvania. That is remarkable. So the fear of cheating came to be. It led to New York City Department of Education to block access in December 2022 followed by a ban the following month. Talk about impact.
Edward Tian, a 22 year old Canadian native and Princeton senior released an app on 3rd Jan 2023 to detect AI generated text. A month after the launch. Within 10 days he had 80,000 users and it was no surprise that Turnitin, the market leading plagiarism detector used by academic institutions around the World was knocking on his door. He had written the app during the 3 day holiday period.
Within months of its launch, hundred of books appeared on Amazon that credited the author or co-author as ChatGPT.
Apps are now appearing that tap into ChatGPT to provide a range of utilities. The app ads are also appearing over social media and messaging platforms targeting content creators and aggregators. They go as far as naming and highlighting existing fee paying popular software made redundant by the introduction of their ChatGPT driven apps.
Here is the lynchpin. It can create and debug codes. When the codes in Codex program were entered into exams for coding together with first year students, it outperformed most of them.
Industry and Competition
ChatGPT is not alone. It will soon be joined by others. Google Bard is also an AI Chatbot in beta testing phase and is expected released any day soon. Microsoft has also announced that its Bing search engine will be driven by AI in time.
Google appeared to have taken a different approach such as the selection of source data and how the application is trained. So do expect differences in the way queries are handled as well as the delivery of results.
I have already mentioned utility apps directly using ChatGPT have started to surfaces in numbers. I expect this to continue.
There is one other area of mounting interest. That is SEO – search engine optimisation. This is extensively used in the online World for website ranking and pages views. It has high commercial value relying solely on search engines and their algorithms. If the search engines lose some of their market share to AI chatbots, it will have a direct impact on SEO effectives. It cannot be applied to AI Chatbots in its present form.
In a nutshell
ChatGPT answers queries, completes essays and writes codes. The output is coherent and contextually correct and delivered in a human like form.
It however has limitations and these can be reasonably overcome by a discerning individual. In the wrong hands and for the unassuming it can be a different outcome.
The developers have incorporated controls to minimise abuse and misuse by stopping inappropriate queries. It waits to be seen if these are indeed adequate. It was no surprise that academics, legislators and the media amongst others have raised concerns. These were around harm to society in a number of ways and the possible loss of nurturing creativity.
Its potential usage universe is huge and to give some guidance, Google stats for 2022 would be helpful. Google is the worlds most visited website with 89 billion visits per month. 8.5 billion searches per day and 84% of Google searchers do search 3 times a day or more.
Despite its rough edges and valid concerns it is a seminal event. A paradigm shift in the world of technology driven information.
Bard and Bing AIs
Update: Today, on 21st March 2023, Google released Bard to the US and the UK. Access is limited to a test group and a waitlist provided. It is also restricted to 18 years old and above. They tagged it as an “Experiment” and made it clear to use it with caution with the possibility of “misinformation” or in AI Chatgroup parlance “hallucination”.
It should be noted that Microsoft’s Bing search has been enhanced in collaboration with Open AI, the creators of ChatGPT. This was released on 7th February 2023 as a limited preview mode for desktops.