Reflections on six months of using ChatGPT

I've been using ChatGPT regularly for nearly six months now, and I wanted to take a step back to reflect on my experiences during this period.

Reference Milestones

I remember the release of Microsoft Encarta on CD-ROM in the early 1990s. Initially priced around ￡400, it soon dropped to about ￡70. I saved up to buy it and spent hours reading articles, listening to sound snippets, and watching videos. The detailed cross-referencing allowed you to start at one of Churchill's famous speeches and finish at the anatomy of a bumblebee, covering thousands of well-written and interesting articles in between. At the time, it felt like Encarta covered every piece of knowledge known to man, but it was, of course, just a tiny subset.

The next significant milestone was the birth and rapid expansion of Wikipedia in 2001. Wikipedia's crowd-sourced knowledge gathering and curation was a radical departure from traditional encyclopaedia publishing. Everyone was invited to contribute, and the knowledge base grew exponentially. Content was constantly updated, covering a range of specialist and obscure topics. For years, Wikipedia was the go-to resource, particularly for scientific and technical information. Even though Google became more prominent for quick answers, Wikipedia articles often appeared in the first couple of search results, further promoting its use. Dedicated volunteers maintained enough trust in the information available.

For two decades, Google dominated while other search engines fell by the wayside. "To Google" became a verb in common English usage.

Then, in November 2022, ChatGPT came along—seemingly out of nowhere.

The Arrival of ChatGPT

When I first "prompted" ChatGPT, I had the same feeling I did when I loaded the Microsoft Encarta CD-ROM for the first time. I started asking Google-like queries across several different topic areas and got specific, fully-formed paragraphs of information back. The answers were thorough and appeared accurate. The training data for the first version of ChatGPT was valid until September 2021, so questions about current affairs, recent popular culture, and general news were off the agenda.

However, everything else was handled with amazing speed and efficiency. ChatGPT not only matched the effectiveness of previous reference tools but exceeded them in several compelling ways. It was possible to enter much more sophisticated questions and specify the complexity, length, and style of the response. For example, you could ask ChatGPT to respond in the style of a university lecturer or a kindergarten teacher.

Previously, certain topics would take hours of research involving multiple Google queries and digesting pages of information. ChatGPT could produce better results in seconds, aggregating information across multiple sources, picking the most relevant information, and presenting it back in fully-formed, easy-to-read paragraphs. Information could be condensed or expanded at will, and you never felt you were getting to the end of its knowledge.

ChatGPT has many more tricks up its sleeve other than reference, but this use case alone made it an essential tool that I would be loath to give up. Eight times out of ten, I now load up ChatGPT rather than Googling for reference and research activities.

Review Capabilities

Electronic word processors quickly replaced typewriters because they allowed writers to edit their work, perform spell checks, and eventually identify grammatical errors. ChatGPT takes things significantly further. You can present it with a previously written article, and it can suggest additions or omissions, improve the writing style, and cross-reference with other information.

ChatGPT can produce content from the simplest of prompts, but I prefer using it to review work that I have largely created myself. This way, my personal style, thoughts, and experiences are preserved but can be improved with suggestions that I can choose to implement or discard. The beauty is that ChatGPT never tires of repeatedly processing the same material or suggesting another improvement. I have found this iteration to be extremely helpful in transforming the mediocre into something more inspiring.

My son recently completed his A-levels, and during his revision period, he was able to run his past paper questions and answers through ChatGPT. The system produced detailed feedback for every answer with suggestions on style, structure, and content. Answers could be cross-referenced against course material, and areas where more revision might be required were identified. The ability for a student to do this independently is invaluable. No one is judging the quality of the response, and the facility is available every hour of the day from anywhere.

There is much debate on whether AI will replace jobs, and I am sure some jobs will be replaced. But I am more interested and excited about the possibilities of people working alongside AI for increased productivity.

Experimentation

Unlike many technology solutions, the "barrier to entry" for a new user using ChatGPT is low. Install the app or browse the website, create a user account, and start prompting. For free. Even the pro upgrade is a relatively low monthly fee. The prompt/response pattern is so accessible but conceals the power and complexity that ChatGPT possesses. Part of the fun and appeal of using ChatGPT is how you can experiment and get instant feedback.

Following the upgrade to version 4.0, you rarely hit the daily usage limits that curtailed activity in earlier versions. So, just prompt away, explore new areas of knowledge, consolidate existing knowledge, generate quizzes, pictures, code, and impersonate famous people or great writers. Branch off into different directions as the mood takes you and dig deeper when you find something new to explore. It is quick, easy, doesn't cost much, and is so much more rewarding than browsing the web. There are no adverts, no clickbait, and no distracting videos of cats doing amusing things with balls of wool.

Data Analysis

Provided the data set is reasonably small (<1MB), a standard ChatGPT can easily cope with analyzing it and performing sophisticated calculations. Data can be assessed for quality, reformatted if necessary, and improved prior to analysis. Analysis commands can be written in plain English rather than complicated formulas, and ChatGPT can even check its own work and generate the same outputs using several different approaches to increase robustness.

Going from an analysis idea to seeing the end results is straightforward and can turn anyone who is mildly computer literate into a data scientist. I have had fun processing football results, stock trading data, property data, and hotel occupancy rates with the ability to visualize the results, generate single-page web user interfaces, and create AI agents along the way.

Summary

ChatGPT now forms part of my day-to-day toolset along with traditional office tools like Excel, Word, and PowerPoint. It can't do everything, but it is particularly good as a "thinking partner" to help augment and improve a wide variety of tasks and content production.