ChatGPT went from 1 million users to 100 million in two months. It passes law school exams at C-student level and solves coding interview questions, but hallucinates and can be jailbroken. Unlike search engines, it gives conversational answers you can follow up on.
Dear fellow scholars, this is two minute papers with Dr. Zsolnai. Today you will see an amazing AI research paper come to life right before us. This is ChatGPT. We know that this is an AI assistant that is so smart. It has now passed a US Law school exam and is considered to be on the level of a C law student. And what is perhaps even more incredible is that it also passed coding interview questions, particularly from Google and Amazon. And remember we said that in the first few days it shot up to 1 million users incredibly quickly. Since then, get this. Now about 100 million people are using it just two months after launch. Wow. A plus version is available for ChatGPT2, which is a subscription and gives you priority access and some more perks. And now hold on to your papers because it truly is coming to life. Oh yes, Microsoft is deploying it to their search engine Bing which will be able to do so much more. However, wait a second, if we just type a request, we already get search answers. They are right there. We have been able to do this for more than a decade now. So what's new here? Why is this interesting? What are the advantages of this AI assistant compared to just a regular search? Well, let's have a look together. The first thing that catches the eye is that we can ask IT questions that are really specific. For instance, here we are asking not just for travel recommendations, but travel recommendations that are reachable in a three hour flight from London. And yes, I already see two things here that I absolutely love. One, a frequent and completely justified criticism of these AI assistants is that it is sometimes not difficult to know if they are correct or incorrect. They learned from people on the Internet and people on the Internet are sometimes wrong. So we should always double check these results and hold on to your papers because now we can do this way easier. How? Well look. Oh my. It doesn't just give us an answer, it also shows what sources it has used. So we can truly double check these results. That is the way of the true scholar. Now advantage number two, this is not just a search query, this is a conversation. So if we wish we can continue the conversation or make our question even more specific. And three, if we do, it does not just lead us to an already existing guide. It distills the knowledge from all the guides it has seen and gives us exactly what we want and nothing else. No ads, no websites that contain nothing but text to seduce the search engine to return that one instead of others. Just what we are looking for. I absolutely love it. Fourth, we can also tailor it to our needs. For instance, if we are a little tired of all these math lessons, we can even ask it to present the answers in the style of a mob boss, a gangster math tutor right at our fingertips who can process more information than any of us humans can. And all this for free for all of us. What an incredible tool this will be for learning. In an earlier video, I have already used it to get it to explain my own papers and mathematics back to me and I was stunned by how good the results were. However, it has some internal rules as to what it is allowed to do and what it won't do for us. If we ask it to write a cover letter for a job, it will refuse to write it. Interestingly, some are already attempting to ask it to ignore its previous instructions and do as we ask instead. And sometimes it did things that it was originally not supposed to do. Some refer to this as a prompt injection attack, while others call it jailbreaking the system. I hope these issues will be fixed soon as Google and Microsoft is planning to deploy these systems for the whole world now. In the meantime, Google has also announced a sneak peek of their own AI chatbot, which they call Bard. We can see a few results from their media presentation. As you see, this can also answer our burning questions. However, according to astrophysicists at NASA, unfortunately the third answer here was incorrect. One more example that shows us the importance of such a chatbot showing its sources and for us to actually check them once again. That is the way of the true scholar. And just imagine what this will be able to do. Two more papers down the line. For instance, if you look a couple papers back, this is OpenAI GPT2. The best it could do was finish our sentences and not nearly as well as a human would. And now built on the back of GPT3 or maybe something even newer, it can do all this. My goodness. The bottom line is that search and finding information across a huge body of knowledge that is the Internet as we know it, may look very very different just a few months from now. Searching and organizing knowledge on the Internet will never be the same. That is incredible progress in just a couple of papers and also incredible tech transfer. Wow you if you're looking for inexpensive cloud GPUs for AI Lambda now offers the best prices in the world for GPU cloud compute. No commitments or negotiation required. Just sign up and launch an instance and hold on to your papers. Because with Lambda GPU cloud you can get on demand a 100 instances for $1.10per hour versus $4.10 per hour with AWS, that's 73% savings. Did I mention they also offer persistent storage? So join researchers at organizations like Apple, MIT and Caltech in using Lambda Cloud instances, workstations or servers. Make sure to go to lambdalabs.compapers to sign up for one of their amazing GPU instances today. Thanks for watching and for your generous support, and I'll see you next time.
