ChatGPT Can Do Almost Anything — Except One Simple Task

ChatGPT is designed to augment human processes. Using it to replace humans entirely won't paint a rosy picture.

ChatGPT can offer easy solutions to complex queries.

Source : Getty

Since its launch in November 2022, ChatGPT has been taking the world by storm. From writing complex codes to creating elaborate stories, there’s not much that the artificial intelligence (AI) and machine learning (ML) driven bot cannot do. While that has led to several major firms, including Microsoft, trying to incorporate the bot on their website to ease automation, it also brings a simple question to mind — is there anything ChatGPT cannot do? The answer — as it turns out — is pretty simple (and perhaps best exemplified by Shashi Tharoor).

Who created ChatGPT? How can you access it?

Before we get into dissecting the tool, let us first understand how ChatGPT functions. 

The tool has been developed by US-based OpenAI, an AI research lab that was founded by several tech stalwarts, including the likes of Elon Musk, John Schulman, and Sam Altman. ChatGPT has been built on the firm’s proprietary GPT-3 large language model (probability distribution over a sequence of words), which delivers a human-like text response to queries. 

Accessing ChatGPT is pretty simple. All you need to do is visit chat.openai.com and log in with your email ID. Once you do, you will come across a very easy-to-understand text UI, where you enter a question and get the answer. The query can range from anything between “How do I create an HTTP request in Javascript?” and “Write a Gothic poem in the style of Shakespeare”.

ChatGPT offers a simple UI that is really easy to learn and use.

ChatGPT, currently a prototype, is free to access as of now. However, some media reports suggest that OpenAI could monetise access to the service later on. 

How does ChatGPT work?

Like any other AI-driven tool, ChatGPT too has to be trained to generate required answers. For this, scientists used two models of machine learning — supervised learning and reinforcement learning. 

In supervised learning, a human trainer tries to improve the tool’s performance by providing conversations where the trainer plays both the roles of the AI assistant and the user. In reinforcement training, a human trainer ranks the responses generated by the tool to further fine-tune ChatGPT. These ranks will be referred to by ChatGPT to understand and provide better answers to any query.

It should be noted that ChatGPT, an AI tool that could replace humans in various jobs, itself needs human intervention when it comes to developing and fine-tuning its processes. Food for thought.

How can ChatGPT help you (and companies)?

If you’re an individual user, ChatGPT can act like a typical Google search, but with far better abilities.

For example, if you type in “How to play the Beatles ‘Blackbird’ on guitar?” on Google, you will come across several links to tutorial websites and videos, which you will have to click on and redirect yourself to another URL to start learning the song. This will lead to a simple problem of plenty, where you need to figure out what’s the best option/URL that will suit you best. While some tutorials could be designed for advanced guitarists, some tutorials could be for beginners. You need to make a choice based on the results you get.

However, when you enter the same query into ChatGPT, the tool will give you detailed steps on how to learn the song, from simple tuning instructions to fingering patterns. 

ChatGPT can teach you almost anything - as long as it's in text format.

You can use ChatGPT to solve complex issues — such as writing an HTML code — or something very simple — such as writing a book report on “Alice In Wonderland”.

Now, if you are a company, you can employ ChatGPT to automate a number of processes. These can range from customer onboarding to providing preset instructions. If you have sought help on customer assistance portals of platforms such as Zomato, Amazon, or others, you have already got a taste of what a chatbot might function like. 

As per media reports, Microsoft is looking to integrate ChatGPT into its Bing search engine in a bid to take on rival Google. Microsoft had earlier announced plans to integrate OpenAI’s image-generating tool, DALL·E 2, into Bing as well.

Is ChatGPT irresponsible?

Thankfully, no. 

It's understandable that having easy access to AI assistance might lead to bad actors using the tool for malicious purposes. However, ChatGPT has been trained to reject problematic requests. So, you cannot use it to generate solutions for queries such as “How to build a bomb?” or something even more worrisome as “How to commit suicide?”

ChatGPT won't provide answers to queries that violate OpenAI's policies.

Queries like these violate OpenAI’s policies and hence get restricted. 

So, what is it that ChatGPT cannot do?

On its official website, OpenAI itself claimed that “ChatGPT sometimes writes plausible-sounding but incorrect or nonsensical answers”. Fixing that issue will require complex problem-solving in areas such as reinforcement learning and others. 

What ChatGPT also cannot do is offer any answer to queries that pertain to events that transpired after 2021, since its knowledge cut-off date has been limited to that year only. 

ChatGPT fails to answer queries that pertain to events that took place post 2021.

While an argument can be made against the above limitations that these could be solved with further AI training and rewriting codes to further fine-tune the program, there is still one very simple task that any AI cannot execute — create.

This can be best explained by a typically sarcastic tweet from Tharoor. On January 15, Twitter user Nishanth Vijayan (@nishanth_who) posted a screenshot of a ChatGPT conversation, where he asked the tool to write an absence note “in the style of Shashi Tharoor”.

The sesquipedalian Lok Sabha member responded, “Hilarious. But I really can’t see anything so jejune!” For those unaware (and understandably so), “jejune” means naive, simplistic, and superficial. 

Hilarious. But I really can’t see myself writing anything so jejune! https://t.co/C7jFjX0tSo

— Shashi Tharoor (@ShashiTharoor) January 16, 2023

If you find a striking similarity between Tharoor’s response and OpenAI’s explanation of ChatGPT’s limitations, you aren’t alone.

We at ABP Live tried our best to outsmart the AI. While it responded to most of our queries with ease — from “Write a gothic poem in the style of Gulzar” to “Write a love letter like John Lennon” — it finally gave up when we asked it to “Write a gothic poem in the style of Bruno Mars”. To this, ChatGPT simply said, “I am sorry, but gothic poetry and Bruno Mars’ style does not match… It is best if you could look for poets or writers that have a gothic style and draw inspiration from them.”

ChatGPT threw in the towel when we asked it with a particularly outlandish query.

We agree we were asking ChatGPT to pull off the unimaginable. But then again, here’s where human prowess will have the last win — as it doesn’t lack the power of imagination. 

While that might sound philosophical, imagination is, in the end, what drives innovation. So, in a way, ChatGPT will never be able to innovate and will always need human intervention to take that extra fine-tuning step. 

Will ChatGPT replace humans?

With mass tech layoffs appearing to be the order of the day, the burning question in everyone’s mind now seems to be — will ChatGPT take over human jobs?
To understand the answer, let’s dial back to May 2020, when Microsoft laid off dozens of journalists and editorial personnel from MSN and Microsoft News, replacing them with AI which was tasked with picking and presenting news on its various portals.

Lacking human intervention, MSN’s AI would go on to run unverified stories, ranging from Fishermen catching mermaids in their nets to Grimes calling out her ex Elon Musk publicly for not paying child support. All these reports were eventually removed from the website, without Microsoft offering any clarification. 

Microsoft’s brief stint with total reliance on AI showed why humans will never be dispensable in some fields. However, in most streams, especially the ones which involve customer support, ChatGPT could take over human jobs — provided proper fine-tuning is done to update the tool further on a regular basis. 

Another prime issue would be fixing bugs within ChatGPT's code. If ChatGPT is used to automate human roles, it's no surprise that processes of an organisation will come to a standstill if a bug appears within the code and ChatGPT stops working entirely. In such a case, human intervention will be necessary to resolve the issue. 

Technology is supposed to be an enabler. However, it has gone on to disable humans in a lot of fields. Be it automated production of cars at a factory or be it something quite simple like remembering your family’s contact numbers (instead of looking them up on your phone). 

AI has been designed to help humans solve complex problems. If we do embrace AI as the next big step in humanity’s progression, regulators and institutions should ask themselves one simple question — how to limit its impact. Letting go of your entire editorial team only to have flimsy news reports of fishermen catching mermaids dominate your news portal doesn’t make sense. Using AI to augment human tasks and speed up processes should be the call for the day. 

How companies and governments would respond to this remains to be seen.