The Versificator is here.. could it eat your job for breakfast or make you do it better?
After the telephone, this could be the next big thing..
n=7;
It is time now to marvel at the imagination of George Orwell, who in his last but most read and praised book about a dystopian world, titled 1984, talks about a device called versificator :
The tune had been haunting London for weeks past. It was one of countless similar songs published for the benefit of the proles by a sub-section of the Music Department. The words of these songs were composed without any human intervention whatever on an instrument known as a versificator.
A figment of one shag tobacco smoking writer’s imagination in late 1940s is almost coming to life. He may have imagined for this to happen in 1984 but either he was betting far too much on the likes of Alan Turing or the book’s title was merely a placeholder for some time in the near future.
Anyhoo, the thing is that this research and deployment organisation called OpenAI (which has some awesome backers - Musk, Reid, Thiel, et al) aims at an arduous goal of developing and promoting friendly AI, whilst the rest of the world is still working on humans to retain that trait.
So they released a paper about GPT3 in May and then the closed-beta api to test-drive.This isn’t the first release, earlier two versions also created a Brouhaha, far too less than this one though and rightly so.
Now, some of you who jump at a new hot buzz word and skim the internet for grasping it’s most talk-worthy aspects to sound aware and smart, would already know that folks who have tried the api,have observed it’s capability to write(generate) prose,poetry and code amongst other things like having a senile conversation as most humans.For the uninitiated or the tech-averse, don't get your greys in a twist yet, it’s awesome and just in case you lose your job to this black magic, you should read along.
So first things first, GPT stands for Generative Pre-trained Transformer (you can get them in a twist now) and it is not an autobot from Cybertron. It is a deep learning powered language model used for NLP (natural language processing) which is another way of saying that it mostly gets what you mean- a trait hard to find in people I guess.
Unlike earlier approaches, the transformer model gobbles up a hunk of text to interpret, translate and summarise rather than serially clanking at it like a typewriter.Feeding data to a neural network / machine learning model is called training and that is a pre-requisite for a model to be smarter. If you lock a kid in a room most of the time, there’s no learning but if you expose her senses to the world and it’s experiences, the brain starts to learn and grow.Similarly, machine learning models ought to ingest, learn and comprehend a lot of data to make connections between words, concepts and contexts.GPT3 went through a lot of reading:
And by the estimates of some smart hardware folks out there, it must have cost about $5 Mn in terms of compute and memory on the cloud, just to do all this.
As a result, the GPT3 language model arguably surpasses the results of all earlier NLP efforts (BERT, GPT2, T5) whilst climbing on those shoulders.
How does it work?
So GPT3 works using next word prediction, the same as it’s predecessor GPT2. Which is like fill in the blanks. I give you an example or more and you are able to guess the missing word in a sentence I give you next.Oversimplification :
1.When the dust settles.
2.When we cross the bridge.
3.When the fat lady ____.
At this point, it is important for us lesser mortals to understand the concepts of shots - zero shot, one shot, few shots learning which reminds me of the t-shirt - 1 tequila, 2 tequila, 3 tequila floor !
Zero shot : without providing any data as reference, classify/identify/respond.Like, you speak a sentence and I tell you which movie is it from.
One shot : you show me one cat and I will recognise all cats out there.And I will do it so well that anytime you see an animal that looks like a cat in the obnoxious re-captcha you encounter because a website thinks you are a machine, you will check with me if it’s a cat
Few shots : You get the drift..
What seems to have been observed by various folks who had access to the api (I’m still on the waitlist), is that it’s smart under the hood.It does some crazy things very well which makes it hard to believe that you are talking to a human and does some simple things really badly to make you believe it’s a human.
Does it work like a charm?
So, it’s not all doomsday for the lot who’s in the lower rungs of knowledge economy because while it might be a superlative step in the direction of Artificial General Intelligence, bereft of smart and worthwhile applications, does it remain a toy to marvel at?Well, the jury is out there - where it usually always is.
Could it eat your job for breakfast?
The obvious applications like customer support, etc come to mind as usual but these are areas that have been talked about earlier as well.Till such time that a language model is able to have twisted conversations like a sober human,replacing jobs is a far cry.Having said that, doing numbing tasks, which has been the endeavour of many a people using NLP would continue. For instance, making FAQs was always there but now it is in the form of interactive interfaces called chat bots.These interfaces might get smarter but not equivalent to a human, not yet.
By the by, the genesis of Artificial General Intelligence stems from the debating materialism v/s dualism philosophies about the concept of mind - the former suggests that the mind is physical and thus leads us to believe in the possibility of it’s replication in parts or whole artificially but the latter denies the existence of any physical properties of the mind.But I digress, we need to talk about the job-snatcher.
So.
1. If you are a lazy or terrible software engineer and want to not even go to stack overflow to copy and paste code, you can buy access to this api and see if it writes all the code you’d like.
2. If you are a publication going under water and can not afford to pay writers and editors (not that they are paid very well anyways), get a load of this api and rehash the publication which still many folks will not understand
3. Want some cool pickup lines that are original, go GPT3 !
4. You're a movie maker or a screenplay writer who wants to be original or write a script drawing from John Cassavetes, Akira Kurosawa, Michael Powell, Satyajit Ray, Martin Scorsese and Guy Ritchie all at once, Go GPT3 !
5. Want to respond to your teenage kids aptly or in their parlance ? Go GPT3 !
6. you’re a writer who would love to get the Booker whilst stirring a literary cocktail oozing J.G Farrell,J.M. Coetze,Salman Rushdie,Paul Scott,Arundhati Roy,Penelope Fitzgerald altogether ? Go GPT3 !
7. You get the drift ..
But, can this technology replace the annoying IVR / email or chat conversations and take them to a new level of irritability?Maybe, maybe not.
Can it be the advent of ux interfaces that make more sense than the confounding registration forms that one has to fill? Who knows?
Can this mark the end of fine arts the way 3-d printing will kill sculpting, artificial (not so general) intelligence will kill music composition or painting? The death of fine arts as we know it ? The Renaissance, Baroque,Impressionism,Realism,etc have all had one goal - to teach algorithms how to paint?The art galleries sure will be bummed - how would they fool the art aficionados now ? Here, this one was made by a 2GPU machine running out of knowhere !
I don’t think so.There’s more to us humans than relying on machines to make us smarter yet lazier.We are doers of er, things.We are an inspired species with each individual born carrying lofty purposes in their underbelly,yet searching for that very purpose of life.We are the evolved ones for whom survival, adaptation and proliferation abated to be the core objectives of life yet we have the switch in our reptilian brains to go back to those instincts any time.
GPT3 is awesome.Artificial General intelligence is amazing and scary at the same time.It is not a human replacer but a skills’ compounder in the making that can dispense certain capabilities to those who do not possess them, at a cost that OpenAI will charge for accessing.
What next - humans making pen pals with machines?
Language, fine arts and above all cognition - these are very human traits.Replicability,replaceability and even superior capability via machines has fascinated us to move towards Singularity,knowingly yet inadvertently.
It’s a commercial release, GPT3, so there will be soon a beeline of companies seeking access to the api and depending on the pricing that OpenAI pegs.These companies would then create use-cases that will spread like wild fire there is a huge possibility that NLP as an ecosystem will expand with developers and companies finding newer and unimaginable ways to use it in our daily lives.Just like the voice assistants are doing - they are everywhere in our lives because ever so often, we need to play a song or stop playing one.
What GPT3 does prove is that machine learning models (context: NLP) can get better and better if we give them a lot of data with a lot of hardware.So obviously, there’s these companies that we call Big Cheque(!) or something, and have fangs.They are certainly going to assemble a handful of their brightest minds, give them the chump change needed for hardware and ask them to get ahead of GPT3.Soon after that, we will see NLP capabilities being made available in some devices and then reasdy-to-use in the cloud for a price which will be cheaper than what I used to pay a certain fellow for renting comic books as a kid.
One important aspect to bear in mind is that,most of the data used for training the language models - from the internet and books is going to mimic our culture with all it’s deficiencies, flaws and biases.A lot of mindful curating of the right kind of data ought to be done therefore to weed out what we are struggling with in real life.There are also tremendous possibilities if misuse - earlier versions were pulled back due to fears of generating fake news, opinions, etc and spreading them.But organisations and people like OpenAI exist today for that very reason - to disallow misuse, mutation and irreversibility of machines’ capability.
Is that enough then?