So How Does ChatGPT really work? Behind the screen!

FOLLOW UP VIDEO:

For Input on Videos and Early Access, Join Me here:

REFERENCES

CHAPTERS:
0:00 What is ChatGPT?
1:33 Magellan offer
2:31 How ChatGPT differs from Google
4:26 Overview of how ChatGPT works
7:07 Simple example of what happens behind the scenes
9:45 Beyond sentence completion
10:21 Three stages of pre-training process
13:24 The huge dataset used

SUMMARY
ChatGPT is an intelligent chatbot that uses natural language processing. The GPT stands for Generative Pre-trained Transformer, which means it generates responses, it is pre-trained by humans, and it transforms input data into an output. This model was created by an artificial intelligence research company called OpenAI.

ChatGPT's power is the ability to interpret the context and meaning of a query and produce a relevant answer in grammatically correct and natural language, based on the information that it has been trained on.

It uses neural networking, with supervised learning and reinforcement learning, two key components of modern machine learning. What it does fundamentally is predict what words, phrases and sentences are likely to be associated with the input made. It then chooses the words and sentences that it deems most likely to be associated with the input. So it attempts to understand your prompt and then output words and sentences that it predicts will best answer your question, based on the data it was trained on.

It also randomizes some outputs so that the answers you get for the same input, will often be different. How ChatGPT fundamentally works, is that it tries to determine what words would most likely be expected after having learned how your input compares to words written on billions of webpages, books, and other data that it has been trained on.

But it’s not like the predictive text on your phone that’s just guessing what the word will be based on the letters it sees. ChatGPT attempts to create fully coherent sentences as a response to any input. And it doesn’t just stop at the sentence level. It’s generating sentences and even paragraphs that could follow your input.

If you ask it complete this sentence, “Quantum mechanics is…” -- The processing that happens behind the scenes goes something like this: It calculates from all the instances of this text, what word comes next, and at what fraction of the time. It doesn’t look literally at text, but it looks for matches in context and meaning.

The end result is that it produces a ranked list of words that might follow, together with their “probabilities.” So it’s calculations might produce something like this for the next word that would follow after the word “is”:

a 4.5%
based 3.8%
fundamentally 3.5%
described 3.2%
many 0.7%
It chooses the next word based on this tanking.

But the sentence completion model is not enough, because you might ask it to do something where that strategy might not be appropriate.

In the first stage of the training process, Human contractors play the role of both a user and the ideal chatbot. Each training consists of a conversation with the goal of training the model to have human-like conversations.

Through this supervised human-taught process, it learns to come up with an output that is more than just sentence completion. It learns patterns about the context and meaning of various inputs so that it can respond appropriately.

But human training has scale limitations. Human trainers could not possibly anticipate all the questions that could ever be asked. For this it uses a third step which is called reinforcement learning. This is a type of unsupervised learning. This process trains the model where no specific output is associated with any given input.

Instead the model is trained to learn the underlying context and patterns in the input data based on its earlier human-taught pretraining.
#chatgpt
This way the model can process a huge amount of data from various sources, and learn the patterns from texts and sentences of a near limitless number of subjects. The dataset used to train ChatGPT which is based on GPT-3.5 is about 45 terabytes of data.

[joeyvico]

I have watched dozens of other videos where programmers, computer scientists, software engineers, etc, attempt to explain how GPT works. Your explanation is the best of all.

[-Seaheart-]

It would be cool to have a follow up video explaining in more detail how it is capable of coding

[williamkacensky]

Yes, please continue with this, Arvin. Thank you.

[picksalot]

More follow-up videos please. This is one of most significant and powerful advances in AI. It is likely to affect most aspects of human life, and more rapidly than we can assimilate and understand. Thanks

[KaushikAdhikari]

Make the follow-up, please! LOVED THIS ONE

[korakys]

Somehow Arvin does it again. It's not just physics he's excellent at teaching, this is the best simple explanation of ChatGPT I've seen.

[snakeb]

Great video. If you do make another video on this topic, I hope you focus on specifically why previous natural language processers could not achieve what chatgpt has

[lutfurrahaman]

The way you explain complex subjects supported by relevant visuals is not only amazing but is also capable of inspiring people to seek for knowledge. I would definitely love to see any follow up video on this topic.

[MarkMichalowski]

This has to be just about the best explanation I've found so far, Arvin! Thanks - and yes, a follow-up would be great.

[reinhardpfaff]

This is one of the best videos about ChatGPT. A sequel is very welcome. Thank you very much.

[benjamindover]

Years ago there was a chatbot you could talk to online. I found it really fascinating. I asked many questions. And it asked me questions. I accused it of being a bot, and it accused me of being a bot. That's when I started to realize I wasn't talking to a bot, but to another person. Both of us thought the other was a bot. But there actually was no bot. The whole think was a cherade. It is interesting to hear that this is how they were training these bots.

[KeithCooper-Albuquerque]

Arvin, great video! I would love follow-up videos on the inner-working of ChatGPT and also a primer on any of its coding algorithms.

[yogiwp_]

Fascinating. Would love a follow-up video! Very interested in how it does general reasoning, like solving logic puzzles or trick questions that are not in the training data.

[hallos]

More follow-up videos, please. Your explanations are excellent. It would be nice to dig deeper into the technical aspects.

[SteveGouldinSpain]

I was using ChatGTP to code a screen-scraper in Python for a dataset on a specific url. It tried and failed several times, saying the issue was down to the dynamic nature of the website. Then it really blew my mind. It actually suggested using an alternative website which contained the same data, which I didn't know existed, but it recoded the screen scraper with the new url and it worked! This was using the GTP4 model. It's scary how smart it is. I'll give you a buffalo nickle if you can tell me how it did that!

[BRL]

Yes, make a video where you explain in depth how neural networks work and the math behind it. Thanks

[ReizarfEgroeg]

ChatGPT is a computer program that can understand human language and generate responses to questions or statements. It works by processing text input through a pre-trained neural network called a transformer. This network has been trained on a massive amount of text data to understand the structure and patterns of language.

When a user interacts with ChatGPT, the system uses the pre-trained transformer to analyze the input and generate a response based on its understanding of natural language. The transformer is made up of multiple layers that can process sequences of text in parallel, and it uses attention mechanisms to focus on important information. The output of the transformer is a probability distribution over the next word in the sequence, and this process is repeated iteratively until a complete response is generated.

Overall, ChatGPT is a highly advanced language processing system that can generate human-like responses to a wide range of text inputs.

[Snowflake_tv]

It's awesome, Arvin! I'm looking forward to the follow-up-video!

[GaryGoldstein]

Yes. Please make the follow up videos. Thank you.

[emergentform]

This is incredible! So simple and yet so powerful. Great vid Arvin hooray!