[Reading time: 3 minutes]
Okay, try entering “Tell me the best places for Sushi in Mumbai” into ChatGPT, and see the response.
Once you’ve read it, type the exact same question again, and see the response.
If you’ve played around enough with ChatGPT, you’ll know that both times, it will give a different response, even though both questions are exactly the same.
Why does this happen? Why does ChatGPT not give the same answer for the exact same question?
Well, it essentially has to do with how ChatGPT’s model works, down to its core.
Stay with me in this article, and I’ll tell you why this happens.
I promise this will be one of the most interesting things you read today (it was for me, at least :D)
You see, when we ask ChatGPT to give us some output, at its fundamental level, it works by trying to produce a “reasonable continuation” of whatever text it has got so far.
This means that it tries to write a sentence after reading (and training itself on) billions of sentences, and trying to gauge which word should come next, after the current word.
So for example, when ChatGPT does something like writing an essay, what it’s essentially doing is just asking over and over again “given the text so far, what should the next word be?” - and each time adding a word.
Now, it doesn’t have one word to choose from. It has a list of words, along with the probability of each word appearing after the current word (which it has from its training data).
And from this list, it chooses a word to add.
By logic, we’d think that the word it adds is the one with the highest probability. But this is where the magic happens.
If it picks the “most probable” word, it will give an expected result which will be boring and predictable. Every time.
This is where the “creativity” gets lost.
So what ChatGPT is programmed to do instead, is add a bit of randomness.
What it does is, it doesn’t pick the highest-probability word every time. Sometimes, it picks a word that ranks lower in the list. Randomly.
And this randomness adds creativity to the content.
Now because this is literally “random”, you will never get the same output for the exact same prompt given to ChatGPT.
It’s like two writers - No matter how brilliant they both are, they will never write the exact same book, word for word. Because the thought process of each writer is different.
Every output of ChatGPT is therefore, like a different writer.
Oh, and this is not just ChatGPT. A lot of the GenAI models you see work on the same principle - it’s called a “Probabilistic AI model”
Fascinating, no?
That’s it for today. I’ve kept this newsletter short since this is a super interesting concept and forms the base of everything in AI. Don’t want you to get confused :)
Do let me know if you’d like to learn about any other topic on AI. Just hit reply and let me know, and I’ll write about it!
(If I don’t know the topic myself, it may take some time for me to learn, but I’ll write it for sure!)
And don’t forget to share this if you think you learnt something new :)
Something cool I explored in AI last week
I tried this tool called lovable.dev last week. It claims to work like a fullstack engineer. And I think it does a pretty good job at it - I designed an entire landing page for a website with this tool, in less than 15 minutes!
Content, design and hosting - all in 15 minutes. Here’s the sample page I made (just a demo, so the links won’t open).
This tool works like a charm!
Please do give it a shot, and let me know how you found it!
Couple of other things I’d like to ask (and tell)
As you know, this is a brand new newsletter (and this is only the second email). So I’d love if you could tell me what you’re looking for, from this newsletter. It’ll help me write relevant content that will eventually help both of us. Please let me know HERE - Just 3 mandatory questions, won’t take more than a minute!
I also write a newsletter simplifying business and money management. It has more than 10,400 subscribers as of today. If you’d like to read more about startups, the economy or money management, subscribe to it here
I write smaller, more frequent content on LinkedIn - You can follow me there for byte-sized stuff
If you liked this piece, please share it with your friends (SHARE button below). It’ll help spread awareness faster :)
Until next time…