What Is The Hype About?
The launch of Generative Pre-trained Transformer 3 (GPT-3) has created quite a sensation in the AI world for the last 9 months.
What Is It?
As per Wikipedia, Generative Pre-trained Transformer is a deep neural network that uses the attention mechanism to predict the next word in a sentence. It is trained on a corpus of over 1 billion words and can generate text at character level accuracy.
“Predicts the next word of a sentence given the previous words in the sentence.”
It can understand the context !!
Natural Language Generation (NLG) field of NLP
Where Is All This Getting Applied?
There is a multitude of possibilities:
- Content Creation: Digital Marketing, generating content for marketing campaigns, e.g., Copysmith.
- BlogMaker: Generating a blog on any topic with GPT-3, e.g., Blog Idea Generator – gives you a full-fledged blog with facts and figures as comprehensive as those made by humans.
Do you know? GPT-3 wrote its own blog. Here is a glimpse of it.
- Explain Code: It could be the next digital tutor that explains the code – code -> English translation
- WebSite Creator: Create a website from English description – check out Debuild.co.
And a lot more!!
Brief History and Quick Facts
- Model built by OpenAI (founder: Elon Musk). Its predecessor is GPT-2.
- With 175 billion parameters, GPT-3 thus far is the most powerful text-based AI created with $4.6M spent on 45TB data as the training data cost.
- Surprisingly good in producing human-like text and more.
- Currently is owned by Microsoft – 10X powerful than the MS-owned Turing Model.
- It plans to power Azure apps and shift them from low-code to true no-code.
- GitHub already introduced CoPilot on top of GPT-3 that assists developers with code snippets.
What Is So Special?
- The performance of models depends on multiple factors like how they are updated/fine-tuned etc.
- With GPT-3, models can perform without fine-tuning, gradient, or param updates, making it TaskAgnostic.
- It can learn with just a few examples.
Note: The following figure captures three settings in which GPT-3 can translate English to French. These can be also tailored to suit the other applicable case studies.
- Implementing GPT-3 has been purposefully made easy, making it one HTTPS API call away:-
- One core endpoint interaction.
- Adds a huge benefit – with DevOps, coders, IDEs, and Low-code, it can be plugged anywhere to deliver its full potential.
- The DaVinci model is one of the best OpenAIs to invoke amongst a few others.
How to Get Started?
It takes up to 5 minutes to deliver!! Get hands-on: Always yields the best possible recommended route.
Ingredients to cook a recipe with GPT-3:
1. Your favorite text editor, Python, pip, and OpenAI. Request a beta license directly from them
2. Command line – displayed shortly after
3. And most importantly, a knack to learn 😊
For more examples, please check out https://beta.openai.com/docs/developer-quickstart.
GPT-3 is a versatile Swiss knife offering a whole world of new possibilities.
- You do not need task-specific datasets to accomplish a task using GPT-3. It is “Task-Agnostic”.
- Usage is quite simple – No need to download or re-train the model. All you need is an API key to invoke and get results.
- Development of GPT-4 is in progress. It covers more parameters, and the whole world is watching and waiting. To compete, Google Brain trained a model with a trillion parameters. It gives us a glimpse of the plethora of interesting opportunities coming up.
- Its ease-of-use and wide application have caught the attention of the developer and user communities, and many websites and startups are already leveraging it.
- It gives a huge boost in Application Development Acceleration with developers focused more on logic. Microsoft CoPilot is one such example.
- Non-developers can develop websites using GPT-3 with limited proficiency (like the Figma plugin). The user would only need to type the website category they want the tool to generate. So, “give me a travel app” – helps develop the app in a short time.
GPT-3 is great!! But again, every new technology comes with limitations and shortcomings.
- Lack of semantic understanding, Bias in Generated Text (still risky for automated content on social media – a racist blog can be churned out and floated if automated. So human intervention through the manual review is necessary for content, especially the blogs)
- As per sources: Even OpenAI admits its API models exhibit biases in the GPT-3 papers that are often visible in the generated text. As the model is trained on the world wide web, it reflects and stimulates the prejudices and parochialism of its audience online, and those views can be crude and entitled at times. Are you planning to open an automated news agency? Unsupervised/un-reviewed automated content can even put those involved behind bars!!
- No technology is perfect. Especially in the field of AI/ML. With rapid advancements, we are course-correcting and evolving as we proceed.
- GPT-3 is yet to access the non-public data, which could affect the effectiveness of the document types that are typically not public.
- The GPT-3 model does not know about covid yet (not vaccinated 😊).
With all of these arsenals, GPT-3 is creating waves in the community, and as a technical person, I am interested in leveraging the good!!
All texts, codes, and icons provided here are just examples and in no way contain customer data or vendor information. The code used in this article is from the OpenAI site and for educational purposes only. Any resemblance is purely coincidental, unintentional.