The Most Ingenious Uses of ChatGPT

It turns out that ChatGPT is good at taking tests. Indeed, it can pass US law school exams (albeit at the bottom of the class) and is good enough to pass the Google coding interview for a Level 3 engineer with a hefty annual salary of USD183K.

Don’t try it in Asia though, going by attempts to test it with math questions from Singapore’s PSLE national exams – typically taken by 12-year-olds – to ChatGPT. Spoiler: It didn’t go well, and was only able to solve the most rudimentary of questions.

However, you are probably tired of broad proclamations about how AI will disrupt industries and put millions out of work by now. While there is no doubting the disruptive nature of ChatGPT, what are some practical ways it is making a difference today?

To answer this question, I started taking notes over the last few weeks. Below are some examples that caught my attention.

Work productivity

Over in Singapore, the Government is eyeing the use of AI to ramp up its productivity. As reported by the Straits Times earlier this week, a team from the Open Government Products (OGP) has integrated ChatGPT into Microsoft Word over the last month. The plan is to give some 90,000 civil servants access to the service so that they can improve their productivity by speeding up mundane tasks such as the writing of sample e-mails or even crafting speeches.

The service will be progressively rolled out across agencies, starting from the Smart Nation and Digital Government Office (SNDGO). Hosted on Azure OpenAI, it is understood that an agreement has been made to ensure that data from the Government is kept segregated and confidential.

For inner healing

Elsewhere, artist and scientist Michelle Huang decided to train an AI chatbot using the OpenAI GPT-3 Playground tool to create an “inner child” chatbot as an experiment. With journals stretching back a decade, she selected 40 written when she was aged between 7 and 18, feeding them into the AI model to further train it.

She was taken aback with the result: “That interaction felt very similar to a normal texting conversation as if I were texting my past self in real-time. It felt like I was reaching through a time portal, disguised as a [chat box]… These interactions really elucidated the healing potential of this medium: of being able to send love back into the past, as well as receive love back from a younger self, of finding closure with past guilt.”

Lowering the barrier to coding

We know that ChatGPT is good at programming. But it shouldn’t matter to us since most of us are not coders, right? Well, it appears that AI can significantly lower the barrier for even non-coders to perform tasks that require programmers. Take Jacob Ferus, who (presumably) got tired of adding to do items one at a time.

He promptly wrote a service that allows him to dump chunks of instructions for the AI to sort and add as new to do entries. The entirety of the code, or instructions to ChatGPT is all of 16 lines, written in English. It reads like this: “Your task is to parse the notes below and output a single JSON array, nothing else, of items containing information about different parts of the notes where information has been extracted, separated, and categorized. Information that should be added: …”

Or take Ken Van Haren, who had some questions regarding startups that he would like answered. Rather than painstakingly writing the requisite SQL queries by hand to query the Crunchbase database, he decided to pass the job to ChatGPT. Some questions he posed, and which got answered include:

• How many seed deals were done in 2022?
• What’s the trend in total dollars raised for series A deals between 2021 and 2022?
• Which region had the largest change year over year in view of dollars invested in 2022 vs 2021?
• Who are the largest biotech investors in 2022?

No SQL required. You get the gist.

For data science

Finally, even data analysts or data scientists can benefit from AI. On Twitter, Shubhro Saha demonstrated a mind-boggling Google Sheets add-on that he created to automate mundane work in a spreadsheet. In a short video snippet, he showed it completing advanced queries without the need for regular expressions or function calls.

Instead, he typed in queries such as “full state name” to extract the state name from non-standard addresses, and “three-bullet summary, each a few words” to summarize paragraphs of customer feedback.