Let’s learn about ChatGPT, its brief history, and my experience with its free and paid version.
During the onset of the 21st century, I could only imagine robots whenever the term “artificial intelligence” flashed through my mind. That was also around the time I was exposed to some Hollywood titles (The Terminator, The Matrix, etc.) depicting the possibilities of an AI intervention on this planet.
Most of the cinema, owing to their very nature, tried to point out some spicy, apocalyptic kind of situation in which AI tried to enslave humankind. It indeed was entertaining and scary. But in real life, I was highly skeptical about AI capabilities until I had my hands on ChatGPT.
What is ChatGPT?
To a five-year-old, it’s simply an (AI) chatbot that can write essays, solve maths, explain science, talk history, admit its mistakes, point out yours, and do a few more creative things like poetry!
Technically, it’s a Generative Pre-trained Transformer (GPT) trained via Reinforcement Learning from Human Feedback (RLHF) to reply conversationally. It simply means OpenAI (ChatGPT’s parent company) used human AI trainers and a ton of data to help ChatGPT respond like us.
Behind the scenes, it’s based on GPT, which is at its fourth iteration as of this writing. And except for ChatGPT, many tools (like Duolingo, Stripe, etc.) take benefit of GPT’s capabilities. GPT, at its core, is a Large Language Model (LLM), a massive neural network developed for Natural Language Processing (NLP) tasks. Put simply, it helps the machine talk like humans.
If you’re interested, the next session shares some historical facts about it. Finally, I’ll do a hands-on where you can see ChatGPT in action.
ChatGPT: A Brief History
It all started with Google’s development of transformer architecture, which they published in a paper titled Attention Is All You Need in June 2017. Its key advantage over previous AI language models was the ability to relate the importance of words, irrespective of their position in a sentence. In addition, it was quicker, more scalable, more accurate, and more creative than the current AI abilities of that time.
This gave founding blocks for GPT-1, which OpenAI released in February 2018 for research and experimentation. It had 110 million parameters, signifying its vast ability to handle complex tasks.
In February 2019, OpenAI developed GPT-2, having 1.5 billion parameters. Among other things, it increased its ability to handle even more complex queries, and it could keep a longer context in a conversation.
However, GPT-2 was riddled with its own issues. It sometimes got diverted from the intended prompt to produce offensive content, a result of the biases present in its training data. Moreover, long conversations could easily lead to incoherent responses.
OpenAI tried to address these and more problems with its third update, GPT-3, in June 2020. With 175 billion parameters, this was a massive update over its predecessors. Still, it suffered a few GPT-2 problems, like producing nonsensical text.
Finally, OpenAI took curtains off the GPT-3.5 AI chatbot, ChatGPT, on November 30, 2022, for free public testing. Taking note of its popularity, it soon launched its paid version with faster response rates. At that time, it was the fastest product to reach 100 million active users, taking just two months after its launch.
Currently, we are at ChatGPT Plus (GPT-4), which is reserved for its paid subscribers, whereas the free version is based on GPT-3.5.
So, this was a glimpse of what took us here. If you’re interested in exploring the ChatGPT history in depth, I would suggest checking the official OpenAI website for more information. Alternatively, sign up for ChatGPT and ask itself!
Before proceeding, we should note that no AI tool, ChatGPT included, is perfect. You should always check its output for accuracy. Besides, ChatGPT had a data cut-off in September 2021. This means its responses aren’t the latest unless you use ChatGPT plugins with its paid version.
How to Use ChatGPT?
Register for an OpenAI account. You’ll need an email or can use accounts at Google, Microsoft, and Apple to sign up.
This is the interface:
The left side hosts the prompt history, and the center is where you choose the models (GPT 3.5 or 4). Since this is a chatbot, one simply prompts in the Send a message empty field and press enter for the response.
While GPT-3.5 is good for entry-level things, GPT-4 is for intricate tasks needing a greater level of creativity and reasoning.
For more clarity, we will see how it performs for a few prompts with its free and paid versions.
#1. Travel Guide
I am in London, and I want to visit water parks. Can you suggest some places? And please mention any nearby tourist attractions.
#2. Writing Email
Please write an email to my supervisor requesting five days leave starting the first Monday of next month. Mention my upcoming vacation and the fact that Mr. Johnson has agreed to supervise my projects in the meantime. I'll be waiting for approval before booking tickets.
Summarize the cell phone development timeline.
These were a few prompts to get you started. Now, moving on to GPT-4 for more creativity.
ChatGPT Plus (GPT-4)
Propose an alternative ending to the harry potter saga where Ron defeats Voldemort using a special wand. This wand is nothing but Harry Potter himself, who has turned into this new form for just an hour.
Okay, write a poem about this alternative ending.
*This wasn’t a standalone task but a re-prompt.
Can you guess the next number in this series 23, 95, 479, 2879,…
Just out of curiosity, I have thrown the same prompt to the free version of ChatGPT (GPT-3.5) to see if it can handle this.
See! It failed miserably. It couldn’t follow the pattern and responded completely incorrectly.
This confirms the fact that GPT-3.5 isn’t suitable for prompts demanding high levels of complexity. Still, GPT-4 also has a level beyond which it may reply inaccurately.
This fact highlights a core issue in such generative AI that they are, after all, machines that can’t say NO, even if a specific task simply exceeds their abilities. And that they will answer, right or wrong, without any regard for the possible consequences.
Anyway, I will conclude the prompting part. However, you can tune in to our extensive Geekflare ChatGPT library, where we have done pretty much “everything” possible with this conversational AI.
Currently, ChatGPT is priced at $20 per month for its Plus subscription.
In addition to this chat interface, developers and businesses can leverage OpenAI API to build their AI applications, billed separately. OpenAI also allows fine-tuning its AI models with customer-provided data for tailored applications.
Future of AI
ChatGPT is becoming a fantastic assistant and doing mundane things in a flash. People have more time on their hands, which is really great for most professionals.
On the negative side, it’s taking up jobs and will take even more in the future with its updates. At present, the most affected domain is copywriting, scriptwriting, and every domain which relies chiefly on creating text.
But this doesn’t end here at white-collar jobs.
How about if AI-heavy companies like OpenAI decide to put their powerful AI models in robots, something like this:
Or if these chatbots become sentient, as suspected by Blake Lemoine, a senior software engineer whom Google fired because of his revelations about their AI having feelings.
Hitesh works as a senior writer at Geekflare and dabbles in cybersecurity, productivity, games, and marketing. Besides, he holds master’s in transportation engineering. His free time is mostly about playing with his son, reading, or lying… read more
Rashmi has over 7 years of expertise in content management, SEO, and data research, making her a highly experienced professional. She has a solid academic background and has done her bachelor’s and master’s degree in computer applications…. read more