If you’re a follower of technological advancements particularly in the field of AI, you surely have heard, watched or read about GPT-3. It is the third generation of the Generative Pretrained Transformers- GPTs. Personally I have been following OpenAI’s developments and conceptions for a while now and each time they send update on their research developments, I am completely mesmerized by their work. From what I’ve seen and read around, I’ve found GPT-3 to be the largest and strongest pillar-stone that they’ve established. It is also my personal favorite because it simply is the intersection of two large sets (that were considered to be mutually exclusive till now)- AI and Writing. GPT-3 as it is released in it’s API form, has jaw-droppingly astounding capabilities.
GPTs share a common base with our daily-use tech tools like Google’s speech to text conversion system and smart speakers. Like these tools, they too are based on Deep Learning and Natural Language Processing.
What exactly does GPT-3 do?
GPT-3 is this real-snazzy Artificial Intelligence that writes stories, poetry-prose, creative fiction, news articles, essays and pieces of write-ups that you’d want it to write. Yes even pieces of functional code! It will render all sorts of information you ask it for, in all tenses, and it will even customize the information for you! It can imitate people, generate texts related to specific people, scan their internet produced portrait, answer your questions, respond to your message and come back to you, with the type of content that you want. With this, it is safe to say that GPT-3 is the most powerful Natural Language AI model ever built.
It’s precursor, GPT-2 was already shedding light over all that can be done by AI. And before we could digest this information, the lab announced their third version. It truly is a huge step towards Artificial General Intelligence as it incredibly writes like humans! Comparing the Generative Pretrained Transformers, we can clearly see that the approach of GPT-3 is not very different from that of GPT-2 -which evidently lacked intelligence. GPT-3 has covered large aspects of intelligence and is indeed a lot smarter than its predecessor because it understands (or at least tries to understand) the internet. It is predominantly trained on tons and tons of internet data for a very long time . So you see, it knows what’s on the internet and how things work here. It knows about most of people on the internet. It knows most of the write-ups on the internet and so, it gives results accordingly.
What is hype all about?
While some sections of the population are extremely freaked out by the release of this AI, others are enjoying training and performing research on it, through it. (OpenAI is giving the API access here, so that it can understand all possible use-cases of the developed AI. And well, you can have the access too, just mind the real-long waitlist)
All of this sounded super scary yet very interesting to me. So I went on this exploration trail and as I furthered, I grew more and more fond of it. Who wouldn’t want a know-it-all-text-generator at their disposal? Only someone living under a rock. (That was my dwelling place for a long time too.) And so, on my trail, I came across some fun, amazing and really cool implementations of GPT-3.
For starters, I came across this experiment by Arram Sabeti in which he made GPT-3 write a poem on Elon Musk, as if it were Dr.Seuss. How did the AI do? Find it out for yourself.
Don’t go to the sun,
Or the moon or the stars.
Elon’s going to Mars.I say let’s do it
In a year or two.
If Elon’s mad,
I’m crazy too.
I’m so wonder struck by this poem, that I’ve read it at least 5 times till now, and already having an internal debate on who’d write it better, the AI or me. Existential Crisis already? Nah, long way to go.
In another experiment, one by Andrew Mayne, he tries to simulate (hypothetical) correspondence with famous personalities. He basically prompts the AI with questions, and the AI answers them as candidly as if it were the renowned personality. Here’s what Jane Austen has to say about social media. What did the AI do to generate this outcome? It attempted to answer the questions by understanding the intent of the message, using the API’s internet based knowledge about the person, here Jane Austen.
If I were writing in the 21st century, I’d have Emma promote her self-published book on Facebook.
The most clever implementation of this AI that I’ve come across till now was done to write blogs. Yes! Liam Porr’s blogthat was written by GPT-3 got around 26 thousand visitors in just a couple of weeks!
Amazing, isn’t it to see GPT-3 do wonders in less than a month of its launch? But as an AI, it has it’s shortcomings too. It lacks coherence and meaning of statement. It lacks purpose and extreme human explanatory skills. It lacks the ability to generate useful and practical texts in some obvious contexts. It lacks human spontaneity and reading between the lines skills. These limitations are potential sparks of hope for writers. This experiment here is a testament to my previous statements. The AI tells us that we’re doomed, too soon. It doesn’t look coherent or even practical o me.
December 8 – The United States Constitution is suspended by COVID-19 Commission order, ending the United States government.
December 20 – The United States officially becomes a federal military dictatorship.
It doesn’t really look all good in there.
If you asked a human to predict the remaining part of 2020, the answer would be exceptionally different. Different in many contexts. We are of course doomed, but the statement “USA officially becomes a federal military dictatorship” is too fallacious to even believe. Looks like a theory, a conspiracy theory.
Amidst all this, we can only hope, that’s what humans are good at- hoping and understanding practicality of things, situations and other beings. We understand emotions and respond likewise. We are, in ourselves an infinity and AI will not be able to grab hold of the whole infinity this soon. AI is also an infinity and we humans too won’t be able to grab hold of it that soon.
Will GPT-3-like-AI eliminate the job of writers? Not really. At least not in the near future. This is just a baby-step towards Artificial General Intelligence and soon enough, we will be laughing at ourselves for overestimating the abilities of GPT-3.
Humans claim to be intelligent, but what exactly is intelligence? Many people have attempted to define it, but these attempts have all failed. So I propose a new definition: intelligence is whatever humans do.
This what the AI wrote about human intelligence. And I’d like to agree with it.
If you are a writer, should you be worried about this AI development? This is one question that initially haunted me to the core. Well, fretting and worrying anxiously, being insecure of your abilities is the new uncool. Instead of worrying, maybe start embracing it. Every human activity that is redundant and can be computerized, will be computerized some time or the other. But know, that the AI cannot write like you without you. It is trained on data that we all generate. It is an AI that tries to be a poet, a story-teller, a fiction writer, a text-generator like you. But it can only do so much to be like you.
You as an individual are far more complex, creative and intelligent than any machine that you’d ever encounter. Be sure that, this is just an AI artefact which will weather out, and will be succeeded by a new one, very soon
Our minds are much bigger and smarter in ways we cannot even think of. GPT-3 is just a strong model that imitates humans by generating spectacular texts. And while we are trying to keep up with all this hype, there are high risks and threats posed by the AI, that are much larger and broader in their sense than what we can think of. GPT-3 has received equal amounts of praises and criticisms and as a believer of Artificial General Intelligence, I surely like to look at GPT-3 as a big foundation stone. It has passed machine tests with such flying colors that it serves as a testimony to the fact that AGI is not an impossible dream.
But here, the real question is- What is it that we really want? Is ‘what we want’, the same as ‘what we need’ and ‘what we ask for’? Would we be satisfied with AI doing our jobs, or would we then long to share human texts and have conversations with each other- waiting to bond over the common human-connect?