Here’s How ChatGPT is Changing The World!

ChatGPT is changing the world as we know it. And today we are going to see how it can give these incredible robots a voice. You will also be able to carry it with you all day with this AI pin and answer your questions on the go. Or you can ask it to get this to also assemble insanely complex robotic cows. And it might even be your doctor, or at the very least help your doctor. But wait, what do these robots have to do with ChatGPT? Well, ChatGPT is a helpful AI assistant and since it now understands your speech and can speak itself thus, it can give a voice to these robots. What does that mean? Well now it can help you and answer questions and even show you around in this building. And it can do all this in a fancy British accent.

Ah, Mr. Matt. My employment as a tour guide provides great satisfaction. I find the dissemination of knowledge rather rewarding, don't you agree? Now behold the rock pile.

Or it can be an excited tour guide.

Welcome to Boston Dynamics. I am spot your tour guide robot. Let's explore the building together.

Or Josh.

This is where we Showcase our fastest four legged robot which can run up to 19 miles per hour. I hope you're not too slow to keep up with it.

Or a teenager like, isn't it super.

Cool how robots can be involved in so many industries?

These are really authentic. But I hope it is not so authentic that the teenager starts to throw tantrums now. Also, a company called Humane has announced this AI pin which is powered by ChatGPT and it can be your interpreter, Help you find your way around the city with a projected screen and more. You don't even have to use your smartphone for these interactions. And with this, ChatGPT can be with you all day long. It has become an integral part of our lives. It is absolutely everywhere. But what is the deal with these crazy robotic cows? Dear fellow scholars, this is two minute papers with Dr. Well, that is a task from the USA Computing Olympiad. This is a problem where we are supposed to build a robotic cowherd and where the answer to the question requires coding up a computer program. According to the Computing Olympiad, the level of difficulty of this task is insane. In my opinion, it is not that bad. But let's see what ChatGPT thinks. Now let's give it this problem and oh, look at that beauty. It seems to understand the problem and dutifully starts writing the code. Nice. It says the solution is 61, is that correct? Oh boy, it is. Good job little AI. A perhaps insane level problem, but for ChatGPT not a problem at all. Now I wonder, has it seen this problem before? Well, you know what, just for fun, let's ask ChatGPT itself. So have you seen this task before? And it says that it hasn't. Good. But wait, when I ask, would you know if you had already seen it? It says no. That is hilarious. Love it. Matt, is this some new capability or could even a previous version of it perform these calculations? Let's see. This is version 3.5. It is extremely fast. I like it. And not quite. Version 3.5 cannot solve it correctly. This is something that is new capability. Also, neither can Google DeepMind's new Gemini. Not even close. And speaking of amazing new capabilities, look at this. This paper was written as a collaboration with a board certified radiologist. And here GPT4 can do two things really well. First is identifying diseases. Now hold on to your papers, fellow scholars, because in many tasks the rate of mistakes is less than 1% nearly perfect. What? This is as good as the best techniques out there in many cases. And in some cases it is crowned as the new best. One, I am out of words. Wow. And two, it can also create easy to read summaries for the patients. Here, in the majority of cases, it is not only comparable, but its summaries are even preferred when compared to the ones written by a real doctor. And at this point I will add a strong caveat to this. This is just a really brief look. Please know that anything in this area requires really careful evaluation. And whenever they ask a doctor about these findings, there is always something that they point out that may be missing. For instance here, maybe a real doctor has to write up a huge amount of summaries per day. And maybe they are using more abbreviations and shortcuts. An AI never gets tired and can do thousands of these in a matter of seconds. Does this mean that it is better? Not necessarily. The human written one may be just as understandable. So the bottom line is that things are messy in this domain and evaluation is subjective. This shows that ChatGPT is a Swiss army knife. It can do so much. This is perhaps a new kind of intelligence being born right before our eyes, fellow scholars. And we also found out recently that ChatGPT remembers. What does that mean? Well, have a look. This is Claude 2.1, a competing chatbot from Anthropic, and it has a context window of about 200k tokens. Now intuitively, this means that in a single prompt you can shovel in about 500 pages of material. You can chuck a whole textbook in there and talk to it or get it to summarize huge and boring budget reports. That is amazing. However, wait, look at this. Context length is one variable, but being able to recall all this information is also important. And when asking questions about the document, it matters where in this document the information was. For instance, if it was early on, it might be forgotten by the time it reads until the end. On the other hand, ChatGPT's context length is about half of this. But look at all the beautiful green. You see that this is a trade off. A bigger context length may be created, but at the cost of fidelity. Not even ChatGPT is perfect, not even close. But you see that its fidelity is incredible. So ChatGPT remembers and don't forget this is something that was just a paper four years ago and and today it is taking the world by storm. It is now out there in the real world solving real problems that were previously almost impossible and it is now an integral part of our lives. How cool is that? What a time to be alive. If you enjoyed this episode and you wish to see more, consider subscribing and hitting the bell icon. When I was a PhD student, my experiments created lots of data for my deep learning projects. But I was not looking for more data, I was looking for insights. This is what weights and biases does and it is the best. Everyone is using it. Try it out now at WNB Me Papers or click the link in the description below.