AI or Aritifical Intelligence, as I understand it, is a trained machine to produce human-like results. For millennia man has made tools & machines to help him. From spears to pulleys to telescopes to assembly lines to chipsets to complex algorithms.
Necessity is the mother of invention.
Proverb
Every need is the driving force to invent a solution for it. This recent wave around ChatGPT has made a lot of buzz not because it is a new concept but because of its most refined implementation.
We live in the age of information, hence its needs and resulting inventions.
In case we haven't forgotten, we consume information digitally. We do read books but on Kindle. We get the news, but from our favourite email newsletters. We stay connected with our family, friends & peers but mostly on our digital devices. Now, it is okay for us to use these new technologies but this will also bring a radical shift on ways and tools for this new day & age. We scour for information on our handheld devices now more than ever. From directions to a restaurant to shoe prices to reading about a symptom, we have been looking for more and more complex information on a daily basis. Hence the need for better processing of this information and an efficient retreival method for daily use. Gone are the days when we used to look around jumping from webpage to webpage to look for something.
Websites have been around for a long time. Web crawlers scanned websites and search engines produced relevant results based on our search terms (questions). This has been happening since over 2 decades, yes 20+ years have passed since we were publicly able to ask basic to bizzare questions and get directed to a list of webpages answering those questions. Then why are we so amazed with AI now?
Well! We're so fascinated by it now because one of the program 'ChatGPT' responds in a human-like fashion. There's a reason why it is like this, more on that later. But let's put this fascination into perspective first.
Imagine, you go to the most extensive library that contains almost all of world's knowledge. You have a particular question about a topic. You feel overwhelmed by having to read so many books to get a consice answer. You might approach the librarian, taking their guidance as to which books can help you. Now imagine that librarian isn't a human. Say this robot had been tasked to read all the books & now it can understand your query & answer you in a conversational tone, almost like a librarian. That's ChatGPT.
Now how can this bot understand and answer questions? We already discussed the understanding part. We have been doing this for almost 2 decades now. Even today, we spend a good amount of time asking questions to a screen knowing that we'll be given some helpful directions. We ask from most specific to most random things, like What should I do on my trip to Paris? Who's the richest man on Earth? Did we land on the moon? What's the fastest car? Did Zuckerberg study at Harvard? And so on. We assumed that the search engine understands us and our search queries kept getting specific and keywords kept growing a longer tail to a point that they're almost a full fledged qeustion.
The fascinating part is how these AI programs are able to produce results in a conversational, almost human-like manner. The answer is in their language models. These bots are trained to do paraphrasing of existing knowledge and generating outputs.
Here are some of the most notable language models currently being used:
- OpenAI's GPT: OpenAI's Generative Pretrained Transformer 3 is the largest version of the GPT language model series. Although now GPT-4 has been released which is claimed to be more "creative". Last year GPT-3 received widespread attention for its ability to generate human-like text and perform a wide range of language tasks, such as translation, question-answering, and summarization. It was riddled with bugs, inaccurate responses & workarounds to get unsafer responses. Now, OpenAI claims that GPT-4 is "able to solve difficult problems with greater accuracy, thanks to its broader general knowledge and problem solving abilities."
- Google's BERT: BERT, or Bidirectional Encoder Representations from Transformers, is a language model developed by Google that has been trained on a large corpus of text data. BERT has been widely adopted for various natural language processing tasks, including sentiment analysis, named entity recognition, and text classification.
- Microsoft's Turing-NLG: Microsoft's Turing-NLG is a language generation model that has been trained on a diverse range of text sources, including news articles, web pages, and social media. The model has been designed to generate coherent and informative text, making it suitable for applications such as content generation and question-answering.
These are just a few examples of the many language models that are currently being developed and used by AI tech companies. As the field of AI continues to advance, it is likely that we will see even more powerful and sophisticated language models in the future.
Please know that I am not a subject-matter expert on AI or LLMs (large language models) so take this as my opinion. I'd love to know more and update my blog post. So please feel free to share your feedback & suggest corrections. Thanks!