Next Industrial revolution: What is ChatGPT? Will it replace jobs?

This article sheds some light related to the question will ChatGPT or AIs in general start replacing jobs and disturbing the industry as we know it.

The inspiration for this article came sometime in the beginning of January 2023 after my LinkedIn got flooded with ChatGPT posts, apocalyptic Skynet prognoses, and hype about the new industrial revolution and machines replacing humans.

Table of Contents

• What is ChatGPT?
• Can ChatGPT model be updated with newer data?
• Would new GPT models be trained from the beginning?
• How to get started with ChatGPT?
• How to properly use ChatGPT and maximize its potential?
  • ChatGPT prompting
  • ChatGPT limitations
  • The quality of output in AI tools
• Conclusion
  • What can we expect from ChatGPT in future?

What is ChatGPT?

Ok, let’s start with the basics. Simply put, ChatGPT is a language model based on Generative Pre-Trained Transformer (GPT) models developed by OpenAI company. OpenAI company stands behind some other AI driven products, such as DALL·E, an AI-driven art generator. We can use it to ask structured questions and it will generated answers based on the knowledge it has. ChatGPT can not learn itself but it can relate between our questions which feels like having a conversation. However, there are many limitations to this, where most recognizable one is the limited amount of conversation history that it is aware of.

According to ChatGPT itself: It is a language model developed by OpenAI that uses deep learning techniques to generate human-like text. It is trained on a massive dataset of text and is able to generate text that is grammatically correct and semantically meaningful. ChatGPT can be fine-tuned for various language tasks such as question answering, machine translation, text summarization, and so on. It is also capable of understanding context, which allows it to generate more coherent and appropriate responses in certain tasks.

ChatGPT knowledge base dates up to September 2021. OpenAI has released several versions of the GPT model, including GPT, GPT-2, and GPT-3. GPT-3 is the latest and most advanced version of the model. It is the largest language model that has been trained to date, with a capacity of 175 billion parameters.

If you are interested in a deep dive to compare GPT versions, check out this post.

Can ChatGPT model be updated with newer data?

ChatGPT is pre-trained language model and it’s knowledge is based on the data used during the training. It is not capable of “learning”. But OpenAI can create new versions of models like ChatGPT by training them on more recent data, which can provide more accurate information on various topics.

Would new GPT models be trained from the beginning?

When OpenAI wants to create a new version of the model like GPT, they can use the previous version as a starting point and continue to train it on new data. This process is called “fine-tuning.” This allows the model to continue to learn and improve its performance on a specific task, and makes it more efficient as it does not have to start training from scratch. However, there are some limits to the amount of improvement that can be achieved by fine-tuning the model, and sometimes the new version of the model is trained from scratch.

How to get started with ChatGPT?

At the moment when I write this article ChatGPT is a free service. Getting started with ChatGPT is super simple.

1. Go to the https://chat.openai.com/
2. Type your question to the chat prompt and ChatGPT will generate an answer for you
3. ChatGPT will generate fairly unique text to answer your question, based on the knowledge it has

However, keep in mind limitations related to the answers you are getting. ChatGPT creators are stating these limitations themselves:

• ChatGPT may occasionally generate incorrect information
• ChatGPT may occasionally produce harmful instructions or biased content
• ChatGPT has limited knowledge of world and events after 2021

While asking simple questions will get you some Siri level fun, keep in mind that ChatGPT is actually an AI-driven tool. Like any tool, you get the best of it when you know how to use it.

How to properly use ChatGPT and maximize its potential?

ChatGPT can handle various tasks:

• speech and text analysis – what 90% of LinkedIn hype posts are all about
• translations – similar to what Google Translate and DeepL services are already offering
• breakdown and explanation of complex issues
• writing stories and essays – but it also “hallucinates”, “lies” and “creates fake references”
• write code – the quality on code depends of the quality of instructions
• debugging code – but debugging is wide concept, and ChatGPT can not know of business cases unless described

Most important thing is to understand how to prompt it.

ChatGPT prompting

ChatGPT can assume various roles. Commonly known ones are:

• Act as a Linux Terminal
• Act as an {{language, i.e. English}} Translator and Improver
• Act as {{position}} Interviewer
• Act as a {{coding language, i.e. JavaScript}} Console
• Act as an Excel Sheet
• Act as a Travel Guide
• Act as a Plagiarism Checker – again this is limited to the ChatGPT knowledge base
• Act as {{character}} from {{Movie/Book/Anything}}
• Act as an Advertiser
• Act as a Storyteller
• Act as a Football Commentator
• Act as a Stand-up Comedian
• Act as a Motivational Coach
• Act as a Composer
• Act as a Debater
• Act as a Debate Coach
• Act as a Screenwriter
• Act as a Novelist
• Act as a Recruiter
• Act as a Doctor
• Act as a Startup Idea Generator

There is an interesting GitHub repository called awesome-chatgpt-prompt, you might want to check it out.

However, it can be used to help you with the quick research. Lets say you are interested in the online-booking market and you want to know what are the biggest players in that space and to gather some quick information.

Here is an exercise for you: Take your favorite search engine and fact-check this table, i.e. first row for CTO and CEO of Booking Holdings.

ChatGPT limitations

Taken into account the nature of ChatGPT as a language model we can imagine some limitations:

• Depends on large amounts of data: Meaning, it might perform poorly related to the tasks and domains where limited data is available. Also, current version of ChatGPT is limited with knowledge ending with September 2021
• Does not have a common sense: Meaning, it generates responses based on the language model rules and the data available to it. In this sense it has no way to understand the questions and responses the same way humans do.
• Can not related to the person its communicates with: In that way, it has no empathy and it can not understand emotions. Meaning, for the conversations where emotions are important (let’s say customer satisfaction) it would not perform well. Just imagine how you feel when you are connected with the chat-bot instead of getting to talk with the customer support representative that can understand and help you.
• Can provide incorrect answers. For example, StackOverflow users were flooding this platform with AI-generated answers and administrators recognized high-percentage of incorrect answers. This resulted in StackOverflow’s temporary policy that ChatGPT is banned.
• Has potential to lie and fabricate the truth, for example when asked to provide science paper references

The quality of output in AI tools

When we consider ChatGPT or any other AI-driven tool for that matter, it is important to be able to validate the quality of the output. In the case of ChatGPT this might be hard if you are asking factual questions about related to the topic that you do not know much about.

I did a small experiment with an other AI-tool called StockAI to generate nice photo for one article. I wrote a LinkedIn post about it already.

☑️ Multiple LinkedIn posts recommended this AI tool.
☑️ Tool claims to create “AI-generated stock photos in high-quality” and presents beautiful images at their homepage

My search query was “i want a developer talking with his team in a meeting deciding on web api structure”… 🙂 What I got was a result that looks like an Alfred Hitchcock’s movie.

Conclusion

There is no denying that ChatGPT has potential to disturb industry and to provide AI-driven support for various tasks including but not limited to content creation, translation, code generation and information retrieval in general. However, it is evident and even stated by it’s own creators that in the current state ChatGPT is still in the testing phase.

ChatGPT has potential to generate valuable content when used in the correct way. Meaning, we have to understand how to prompt it correctly. We also have to fact check everything generated by the ChatGPT, at least in it’s current phase. Furthermore, we have to understand, not only the opportunities that ChatGPT gives us but also the limitations that come with GPT approach in general.

It is not possible to determine in which way ChatGPT will develop, because it is not only conditioned with technical implementation and social hype around it. It will have economical (read financial) aspects and it will search for the way to position themselves on the market. This will surely change the way ChatGPT integrates and gets customer adoption.

What can we expect from ChatGPT in future?

We can speculate on some trends:

• Firstly, ChatGPT might challenge search engines for the part of the knowledge search market, or it could integrate into existing search engines. Personally, I do not see ChatGPT as direct competition to Google Search Engine for example, because these are different concepts for knowledge discovery, each with pros and cons. Most likely we will see some cohabitation here.
• Secondly, it evolves faster than social education and use policies. We can already see different cases where It was used in fraudulent activities, such as generating realistic phishing emails and easy way to build fake websites.
• Furthermore, ChatGPT will for sure influence education systems and result in changes on how to evaluate and grade academical success. Down this line it will also affect content writing industries but it remains to be seen in what exact way.
• When it comes to software development, I believe that software developers are already using automation and AI-driven tools to help them in daily coding, code quality checks, security checks etc. In this sense, ChatGPT has potential to additionally assist and improve quality of work, but not to disrupt or replace complex software archtecture, requirements gathering, design and development.

Finally, I do believe that AI-driven tools, including ChatGPT, have the potential to change the industry as we know it. However, I do not think that AI will replace human workers. Human workers understanding and using AI-driven tools to their advantage will replace human workers.