What is GPT3 and Why You Should Care
The best time to start planning for a post-GPT world was yesterday. The second best time is now.
Five days ago, OpenAI launched ChatGPT, which enabled users to interact with one of the most advanced AI’s built to date.
Now? A million people are already interacting with ChatGPT. To put that into perspective:
Right there, that should have your attention.
One more anecdote from a personal perspective: I got to share GPT3 with a couple colleagues today. Both were blown away by it and after seeing it for 5-10 minutes echoed Linas’s sentiment above: ChatGPT will reshape everything moving forward.
What is GPT3Chat?
Functionally, ChatGPT is an interface with a chatbot. It looks like this:
You type in the box, you send messages, and you get replies. Simple enough. And anyone can try it (once signing up) here.
The magic though is in what’s under the hood.
ChatGPT is based on GPT3, which is one of the largest machine learning models built to date. To give you a sense:
The model contains 175 billion parameters
It was trained on “most of the internet”: 500 billion words were used to train the model.
For the non-technical audience: GPT3 “read” 500 billion words (most of the internet) and used that to make associations between those words. Those associations are stored in the 175 billion parameters.
This is not dissimilar to how a human would learn: read or gather information and store those associations in our brain. Human brains have about 86 billion neurons, so the comparison is apt, though neurons do more than a single parameter can represent.
That’s not to say that GPT3 is a brain. It is quite clearly artificial, but it is advanced beyond anything you think of when you hear the term “Chatbot”.
Some Examples
Twitter has been alit with people sharing GPT3 examples. And I’ll caveat upfront: when I first saw them I pooh-poohed them as cherrypicked examples. The magic of GPT3 really comes from using it and seeing how consistently it produces salient information.
The implications are broad, but I’ll keep a focus on a few work-related uses here that showcase the flexibility of ChatGPT:
Writing Code: Commented and Explained
Admittedly, the problem I gave it here is decently straightforward, as is the solution provided. ChatGPT is finding the mean and standard deviation of the time series and flagging anything 3 standard deviations outside of it to be an outlier.
But still: the code is clean and commented. You could easily get worse code from a newish engineer for a problem like this. But the comprehension (for lack of a better term) is impressive. It knew what I meant by “timeseries data” and “return a list of outliers”, and it understood the code enough to explain it clearly to me.
It can go more complex too. For example: writing code to sync various cloud services to, say, analyze the sentiment of an image:
Explain a Concept:
ChatGPT really shines in explaining things in a way that almost no human is. Let’s consider a fairly straightforward technical concept like webhooks:
Let’s try statistics:
I have some quibbles with the first level as I don’t know that a five-year-old would get it, but ChatGPT still had the right idea: explaining statistical significance at three progressive levels of sophistication.
Drafting Summary E-mails:
Okay, that’s cool, you might say, but searching the web with Google could get you the same information. And sure, it could, but it returns a deluge to you: 500 articles all with their own take and framed in their own way. This synthesis is really helpful.
And we’re not done! Let’s say we want to dig in a bit more to the third point. Let’s just… ask GPT:
So that’s cool - it remembered the context and gave me a pretty decent framework for assessing the cost structures to assess.
Brainstorming/Marketing
Had ChatGPT been around back in 2018, we almost certainly would have used it to get ideas on something I’m not great at: marketing and content generation:
And if you want to make it a bit wild…
And I could go on. While ChatGPT is distinctly not human, interacting with it is as human-like as anything in tech. It understands your language and context seamlessly - just as another human would. And that’s the thing. That’s why it will transform day-to-day work and day-to-day life.
That’s not all!
When going through this with a few folks earlier today, we seamlessly jumped from asking ChatGPT business or coding questions as we did above to asking it to generate an apple pie recipe with a unique flavor profile to ideas on how to get 3-year-olds to fall asleep better, to having it write screenplays and poetry.
It handles it all and produces legitimately interesting content. Sure, sometimes it falls short either due to it’s self-imposed restrictions (i.e. I asked it to write an algorithm that would crack 256-bit encryption and I got the “I’m sorry, I can’t do that Hal” response) or some of it’s quirks that are sure to get better as the technology advances and more and more information is fed into it.
Why does it matter?
The implications of this technology are profound. Especially as it inevitably gets even better. Namely:
Increased reliance on artificial intelligence: As ChatGPT and other large language models become more advanced and widely used, it is likely that people will become increasingly reliant on AI to perform a variety of tasks, such as generating text, answering questions, and providing advice. This could lead to a shift in the way that people work and interact with technology.
New opportunities for automation: ChatGPT and other large language models have the potential to automate many tasks that currently require human intelligence, such as customer service, content creation, and data analysis. This could lead to the creation of new industries and job opportunities, as well as the displacement of some existing jobs.
Changes in the way we communicate: ChatGPT and other large language models have the ability to generate text that is indistinguishable from human-generated text. This could lead to changes in the way that we communicate with each other, as well as the way that we consume information. It could also raise questions about the authenticity and reliability of information, as it may become difficult to determine whether a given piece of text was generated by a human or an AI.
Advancements in natural language processing: ChatGPT and other large language models are a significant step forward in the field of natural language processing, and their existence could accelerate the development of other AI technologies that rely on natural language processing. This could lead to new applications in fields such as healthcare, education, and finance.
Now that you are done reading those… guess what? I didn’t write any of that. ChatGPT did though:
Tricks aside, ChatGPT is right, even if the ordering is a bit off. Automation via ChatGPT will take a while to iron out the last 3-5%, but it will happen. Communication in written word now has humans and machines on the same level: henceforth you have no way of knowing if words were written by a person or an algorithm. Audio and video are soon to follow. Who, or what, will you trust?
And to jump back to ChatGPT’s first point: reliance. I just lazily had GPT generate my summary for me. Sure, I did it to prove a point, but eventually my brain could learn to lean on ChatGPT for some element of critical thinking as humans already have come to rely on our Smartphones as information repositories. But what then are humans really for?
Is it hyperbolic to say that ChatGPT is a step change in the same way the Internet, Social Media, and the Smartphone were? Maybe. But it sure looks like a true step change that will kick off another acceleration of innovation. Go ahead and give it a try and see what you think!
More Reading
Here are a couple more articles on ChatGPT that are good reads!
Oh, a P.S.
I also had ChatGPT write my introduction to Bizlandia, check that out!
