OpenAI GPT-4 - The Future Is Here!

Dear fellow scholars, this is Two Minute Papers with Dr. Zsolnai. Finally, OpenAI's legendary language model AI GPT4, which also now powers their ChatGPT assistant is here. And my goodness, do they promise a lot of goodies. Advanced reasoning, complex instructions, more creativity. This is going to be an incredible journey. So. So hold on to your papers right now fellow scholars, and get this. It can also accept image inputs. Let's put this one to work immediately. Dear GPT4, imagine that we have these ingredients. What can we make from them? Note that the input is not text anymore, but a photo. That is excellent. And don't forget, if one of these caught your eye, you can also ask follow up questions here on how to exactly make them. Now, earlier we looked at some preliminary results of an IQ test where it scored 147. However, I noted immediately that we should be good fellow scholars and take this with a little bit of skepticism and wait for a more formal examination. And yes, here it is. It finally happened. GPT4 gives us human level performance on a variety of academic benchmarks and they report something absolutely incredible. Look, this is what its previous incarnation in ChatGPT was capable of doing. 10th percentile on the bar exam and now maybe 30, 50. Nope, 90th percentile. This means that the new one is within the top 10% of test takers. That kind of improvement in just one paper is outstanding. And I know you're looking and the answer is, of course we are going to talk about this too in a moment. And this is where being able to process visual information means an incredible breakthrough. It doesn't just mean that it can help us with the ingredients or can explain memes, although that is quite helpful. But it gets better. For instance, look at this one. Yummy. On the AMC Mathematical Exam, GPT4 marked with light green performs so much better than the previous version. However, here is where the magic happens. Look. This exam contains some light visual information and when we add the capability to process this to GPT4, it does so much better. It is rapidly improving to match human level results. We are one step closer to human level intelligence. This work is history in the making and I was also very excited about this. The USA Biology Olympiad exam. This needs next level capabilities because it requires visually inspecting and evaluating electrocardiograms and and in general taking in a lot of visual information and giving correct answers. So how correct was it? What? Are you kidding? 99th percentile better than almost all humans. However, wait a minute, I see light green, not dark green. What does that mean? It means that this is the non vision version of GPT4. But this exam has plenty of images. So what happened? Well, the AI got a little help. The images have been transcribed by a human. I would love to see those transcriptions. I bet it was not an easy job. So it seemingly cannot perform this level of visual inspection yet. But I am certain we will see something like this one or two more papers down the line. Perhaps with GPT5. If that sounds interesting, consider subscribing and hitting the bell icon to not miss it when the next paper drops. Here is also an example that I absolutely loved. Little AI explained the plot of Cinderella and each word has to begin with the next letter in the Alphabet from A to Z. Wow, this one is tough. I have to admit I am not so sure if most people would be able to solve this in a satisfactory manner. And look at that. Flying colors. My goodness. And this is one more really cool example from their official demo. What are you saying? That this crude drawing for a website comes in with some jokes and buttons and it not only creates this website with proper JavaScript, but it also understands that it needs to fill in the jokes. And it does that too. Loving it. So when can we try it? Well, OpenAI deployed GPT4 to their ChatGPT plus subscribers and they also made a waitlist for API access. Now I was lucky enough to get access to it. Thank you so much. And it already helped me remember my discussion with a geologist professor from a few years ago. There were some details I did not remember and now I do. Thank you. So who is using it? The answer is that everyone is using it. It can help us organize a huge knowledge base, or it can help us learn a new language and so much more. And this is stack transfer. So in other words, from paper to product, not in decades, but in a matter of days and in some cases even a matter of hours. What a time to be alive. And if you have been watching two minute papers, you saw the whole history of these works from GPT2 to GPT3 and now GPT4. You saw history unfold right before your eyes. Thank you so much for being on this incredible journey with me. Now I will also note that the paper is really detailed in terms of the evaluation of the results, but a little light on the inner workings of the algorithm itself. I hope we get to know more soon. This episode is brought to you by Anyscale, the company behind Ray, the fastest growing open source framework for scalable AI and scalable Python. Thousands of organizations use Ray, including OpenAI, Uber, Amazon, Spotify, Netflix and more. Ray lets developers iterate faster by providing common infrastructure for scaling, data ingest and pre processing, machine learning, training, deep learning, hyper parameter tuning, model serving, and more, all while integrating seamlessly with with the rest of the machine learning ecosystem. Anyscale is a fully managed RAY platform that allows teams to bring products to market faster by eliminating the need to manage infrastructure and by enabling new AI capabilities. Ray and Anyscale can do recommendation systems, time series, forecasting, document understanding, image processing, industrial automation, and more. Go to anyscale.com papers and try it out today. Our thanks to Anyscale for helping us make better videos for you. Thanks for watching and for your generous support and I'll see you next time.