OpenAI’s new AI chatbot ChatGPT has taken the world by storm. Here are the basics you need to know and the expectations.
Unless you live under a rock, you must have heard about OpenAI’s new AI chatbot ChatGPT (Chat Generative Pre-trained Transformer), which was announced on November 2022. OpenAI is also the pioneer of other revolutionary AI products, such as DALL-E and GPT-3. However, ChatGPT is the most impactful for each individual on this planet.
Summarising texts, writing an email, explaining difficult concepts, or suggesting the best gift ideas are a few items that ChatGPT can do. All you need to do is type your question, and it spits out the information which it “feels” relevant.
As of writing this, ChatGPT is still free to use during this research period with limited capacity. A paid version is also available at 20 USD per month.
Table of Contents
What Is ChatGPT?
OpenAI’s ChatGPT can answer any question you ask. It has been modelled so that sometimes it makes up an answer with its trained “mind”. You may think of it as a Google with a direct answer machine in the form of conversation.
ChatGPT is trained with a vast amount of data collected from the Internet over the course of a few years. The trained dataset includes several social media data as well, including Twitter. Hence, it can learn the tone, language, biases, and whatnot.
However, its current data is up to 2021. Since then, it has not been trained on the current data. So, whatever information you receive might be a little older, depending on your question.
However, it can mock up information sometimes, which is entirely wrong. So be cautious about what you want to do with the information. Furthermore, one of the fundamental differences from the earlier language model is that ChatGPT can remember what you have asked before in a “conversation” or “chat” and the following answers are based on the set of questions you asked earlier.
How ChatGPT Works
At its core, ChatGPT was trained with both machine learning and supervised learning with reinforcement learning techniques. In the first stage of training, humans participated as “humans” and “AI” both to collect the dataset on how humans respond. This trained dataset was fed to the ChatGPT’s language model in the next phase. The reinforced learning kicked in after this phase. In the reinforced learning phase, humans manually rated and raked the answers based on the quality, which again fed back to ChatGPT’s dataset.
So, humans contributed to the training of ChatGPT alongside its own learning based on algorithms. This helped ChatGPT to answer like “humans” instead of just cold sentences and texts. It’s evident that this foundational base can be further converted to AI voice generations, video creations, and many more real-world use cases.
Again, remember it is still in the research phase, and sometimes the answers are biased or factually incorrect.
How to get access to ChatGPT
To use ChatGPT, you need to visit the below official OpenAI website and create an account. It will ask you for a valid email address and phone number. You can also log in using your Google or Microsoft account
Once the account is created, head to the same website to access ChatGPT main interface.
How to Use
The main interface resemblance to normal chat apps. You type your question and hit the enter key. It will start responding with its answers.
On the left side, you should see a conversation manager box where you can save a chat or create a new chat. Saved conversations can be accessed at a later point in time.
Since it’s a “language model”, you don’t need to be formal in asking questions or following grammar. You don’t need to type complete questions. And you can be casual with your question.
What to Ask ChatGPT
Since its release, users have used it at a maximum capacity, forcing OpenAI to adopt much higher hardware requirements. Here are some examples.
When asked about when will be the year of the Linux desktop, it answers below:
As you can see, it answers preety standard way. This is a very basic prompt which you can tell it to further modify, like below.
Furthermore, you can ask about lifestyle, and whatever topic you want. For example, when asked about Coffee, it answers below. You can also tell it to format the answers in a list of tables.
Ultimately, it is the question or prompts that matters. A prompt can be a very complex item. Here’s an example of a complex prompt. If it can’t answer, it can try to guide you on what you can do based on its current knowledge.
A prompt can be super complex. It can be detailed. Here’s one example of a prompt about writing a story. As you can see, you need to be very detailed about what you want.
I asked about our website, and here’s what it thinks:
As you can see, to get what you want, you have to ask the right question. Isn’t it true for life in general?
The beginning of AI’s impact on our day-to-day life
Whoever you are, or whatever you do, the AI will start impacting your day-to-day life in some capacity in the coming days. And it’s just the beginning. That brings to several concerns and questions related to ethics and security. AI models need to be dealt with extreme caution. Humans always try to exploit AI or any technical advancement for personal gains. Think about an AI exploit that learns via a remote server and changes its attack vector at runtime. A difficult time ahead.
That being said, a huge benefit will be there for everyone if ChatGPT remains accessible in some capacity. It might help teachers and students to explain and understand concepts in a better way. In fact, a low-cost terminal with access to text-based ChatGPT can greatly benefit underdeveloped countries in educating children and students. If it is used rightly, it will impact generations for decades.
Please let me know in the comment box how you think it will change your workflow.