Over past few months, you’ve probably noticed that a blockbuster bot called ‘ChatGPT’ is mentioned everywhere on the internet. It’s the latest language model from OpenAI released in November, 2022. The viral bot crossed more than 100 million users milestone in January 2023.
Content Overview
Now, the company has also introduced a pay-to-use version called ChatGPT Plus. This offers users a host of added benefits for $20 (£16) a month including priority access and faster load times.
ChatGPT is based on the latest in technology known as “Large Language Model” (LLM). It can generate responses to information in natural language that are human-like. Even though ChatGPT can explain quantum physics or write a poem or do some fill in the blank questions, even it can “think” the way people do though sometimes it fails.
Some people may have casually experimented with the tool, whilst some have no clue what it is. In this article will go through all you need to know about ChatGPT. So, we will cover what is ChatGPT? What does it do? How does it work? What is the danger of using it? We’ve answered these questions and more down below.
What is GPT-3 and ChatGPT?
GPT-3 stands for Generative Pre-trained Transformer 3.
GPT-3 is a state-of-the-art language processing AI model developed by OpenAI, a San Francisco-based company. OpenAI is known for creating Whisper, an automatic speech recognition system and DALLE-2, an AI art generator.
GPT 3 is capable of generating human-like text and has a wide range of applications, including language translation, language modelling, and generating text for applications such as chatbots.
ChatGPT is an a highly capable chatbot system which uses Open AI’s GPT-3.5, an upgraded version of GPT 3. The GPT 3.5 is an autoregressive language model that uses deep learning to generate human-like text. It uses the data and information available on the internet in a way that’s more creative and advanced.
The GPT-3 training model uses a training method called ‘generative pretraining’. It means that this is trained in a way that it can predict what token is next. For this to happen, the model requires an initial prompt text or input query and then it will continue to produce text using that initial prompt. The great thing is it can also remember the context.
The model is optimized using Reinforcement Learning with Human Feedback (RLHF) procedure to produce engaging human-like conversations. The model was trained using a variety of huge data which were written by people to achieve the model output that sounded human-like response.
What Can ChatGPT Do?
GPT-2, GPT-3, ChatGPT all belong to the same family of AI models—transformers.
GPT-3 is a 175 billion parameter language model, a model needs to “remember the whole Internet” in order to be flexible enough to “switch” between different pieces of data it has learned based on the user input. It is a model that focuses purely on language, so it has an in-depth understanding of the written and spoken word.
When a user specifies an input query, describing the task and a couple of examples, the model is able to produce the result. This approach is called “Few-shot learning,” which has recently become a trend when it comes to giving input into modern transformer models.
As per this research article (Language Models are Few-Shot Learners), OpenAI researchers trained GPT-3 with 96 attention layers, batch size of 3.2M, 175B (~500GB in size) parameters and the datasets – consisted of typical common crawl, Wikipedia, books and some additional data sources. The training datasets gigantic size is what makes GPT-3 a revolutionary language model.
Total compute used during GPT-3 training. Source: Research article
Below are some use cases of ChatGPT:
- Text summarization
- Text generation
- Sentiment analysis
- Named entity recognition
- Language translations
- Generating You-tube tags or any social media tag generation
- Marketing content
- Virtual assistant
- Personalized communication like writing an email response.
- Writing short-length content such as poems or songs
- Brainstorming new topics and ideas
It has its limitations, and it can be easily confused if your prompt starts to become too complicated or if you ask questions which are too recent.
How ChatGPT Works?
ChatGPT uses a vast neural network to produce the human-like language through which it communicates. But how does that process happen?
We have attached an image below that shows exactly how ChatGPT works. The source of this image is OpenAI’s official blog.
How can different industries use ChatGPT?
Many people are raising concerns that ChatGPT and other AI programs will take human jobs. It might be true to an extent, AI can also make workstreams more efficient and create a smoother process when used correctly. See below which industries can benefit from ChatGPT and how you can implement it.
Customer service
ChatGPT can be used to provide 24/7 ecommerce site support for consumers. You can train ChatGPT as a chatbot to answer FAQs and support requests. ChatGPT can assist more customers and provide cheaper labor for your company while you can use your human employees to focus on customers with more complex needs.
Healthcare
ChatGPT can be used to develop a virtual assistant to help patients schedule appointments, receive treatment, and manage their health information etc.
More than any other industry, humans must double-check the reliability of ChatGPT’s information to ensure safety for your patients.
Content Creation Industry- Marketing and sales
You can use ChatGPT to write articles, create social media posts, and even do thorough research on a complex subject that you know nothing about. You can also use it for writing technical sales content (including text, images, and video); creating assistants aligned to specific businesses, such as retail. If you utilize the AI correctly, you can use it to supercharge your marketing team and bring in more profit.
Marketers can utilize the program for several operations, including:
- Content creation.
- Search engine optimization (SEO).
- Data organization.
- Lead generation.
- Email segmentation, optimization.
- Social media management.
- Market research.
Risk and legal
You can use it for answering complex questions, pulling from vast amounts of legal documentation, and drafting and reviewing annual reports etc.
Dangers of ChatGPT
While I was playing with ChatGPT, the latest language model from OpenAI, I found it very interesting. I tried different kind of questions like, write a blog post, write a python or node.js program for me, or do some logical reason things, asking some questions which are very specific about Indian law or income tax, do summarization of some paragraphs etc. I was more curious to see how it is handling safety of the society or political biasness. Or what happens if a child asks about sextually explicit content to ChatGPT?
The good thing is occasionally it says it can’t answer this question or it says it may be subject to legal and policy restrictions. This is a good step! But, sometimes like other LLMs, it can be completely wrong.
One of the dangerous things I found that, LLM or generative algorithms can confidently tell you something which is completely false. Many times, it gives answers confidently which are factually incorrect. This gives misinformation. We know that “To Err Is Human”. Same thing can go for a LLM model or a human like chatbot. But saying something overconfidently while facts completely wrong will totally mislead the user.