OpenAI's GPT-4: Eccentric Genius AI!

Dear fellow scholars, this is two minute papers with Dr. Zsolnai. ChatGPT has just been supercharged with browsing support. I tried it myself too, and I think this changes everything. Well, almost everything, as you will see in a moment. And this is going to be an incredible journey. So hold on to your papers, fellow scholars. You see, normally ChatGPT with GPT 3.5 has read a great deal of the huge body of text on the Internet. Internet and therefore it knows a lot. It can help us with so much. But there is a problem. So what is the problem? Well, it can help us so much if whatever we are looking for happened before September 2021. Anything after that is a no go until today, because now we have live browsing support. Let's see together what this is good for now. Perhaps the world's best two investment experts, Warren Buffett and Charlie Munger, just held their annual meeting where they discuss investing, the economy and more. As always, it is a treasure trove of information. However, look, it is more than seven hours long. So it is time. It is time for what exactly? Well, for asking the new ChatGPT to summarize it for us and see if it is worth the time investment. First it starts browsing, browsing and browsing. And then. Ouch. Unfortunately, it cannot watch videos, which is understandable. Second try. Oh my goodness, it is hallucinating. Calm down little AI. This is not a video of a startup called MindGrasp. AI. It's alright. Now I am wondering, does this startup exist or did it make it up? It can tell us what it is. And I checked. It indeed exists. Third try. It also fails. But wait a minute, this time it fails a little differently. Look, a ray of hope. It can look up transcripts instead and summarize those. Yes please. And let's see. Great. They talked a lot about human nature. That is my favorite. So I'll definitely watch this. We could probably even ask for the appropriate timestamps if something specific caught our eye. So you see, it works, but sometimes it is a bit eccentric. Yes, we have an eccentric genius AI at our disposal. What a time to be alive. But wait, if it can browse, it can also help us with other things. For instance, it can visit our coding projects, have a look at our work and even suggest improvements. You see a ton of new code here, but it is unclear what it has changed and what it is trying to do. No matter, look, it summarizes the nature of the changes and also tells us why it has done what it has done. That is absolutely Amazing. And if the database cutoff is not in September 2021 anymore, yes, we can ask about things that take place today. So, market research time. How did the US interest rates change in the last three months? That is a great and concise summary with citations. Loving it. Now tell me the story of interest rates in the last 10 years. We get a lot of data. I like it. And even better, lots of additional blinking cursors for free. Now let's do some actual market research and ask what the economy looked like last time the US had interest rates like this. And then, oh boy, our whimsical genius is at it again. Then I tried again and then it says that interest rates looked like this last time around 2006 and 07 and it ran out of time before it could make a table with the appropriate data. But it kindly asks if it can try again. Of course. And good job little AI. Bravo. We now have some numbers to compare the two situations. Now, I hope it goes without saying, please do not invest on this data. I just ran this toy experiment to see a little clearer what is going on and where we are in the death cycle. So that was market research and now it is time for research research. Namely, I would love to know if there are real scientific experiments about guitars. Now, what experiment am I looking for? I am looking for something about tonewoods. This is something that has been a subject of intense debate among artists who use electric guitars. You see, manufacturers love to sell you all kinds of woods they use for the guitar body and neck, and they say it makes a great deal of a difference in the sound they make. Perhaps it does. However, we are fellow scholars here, so we would like to see hard evidence of that. But for me, no luck. I don't quite understand why, but I never found a real paper on this topic until today because ChatGPT was able to browse its way into a proper paper that is about exactly that. Oh my. Here is the setup of the paper where they actually done precise measurements on what sounds these guitars make if they are made of different kinds of woods. And the result is. My goodness, can that really be? This seems to suggest that they indeed make a difference, at least in this particular setup. However, wait. We can ask our resident Genius ChatGPT to read the paper and summarize it for us. And then. Ouch. It couldn't read the paper. But it's almost all text and it is right over there. Still nothing. Alright, let's give it the abstract results and discussion parts of the paper as plain text and ask the same question again and bam, there we go. The body and the neck both seem to make a difference, but it still uses a great deal of technical jargon. So let's ask it to summarize it for a 10 year old and I think the result is excellent. It compares the body of the guitar to the body of a car and how it goes. Vroom. I love it. Clearly the comparison has its limits, but still I am very impressed. Also, every study has its own limitations. Let's ask it what those are. I am very excited to see this. So let's see. Oh come on, let me cultivate my addiction. So quite a bit later we get this and I have to say that this is an excellent assessment of the paper, especially that it only read parts of it. These are just two types of wood and in laboratory conditions and a bunch more great insights. I absolutely love this. So this is the story of our little eccentric genius. It can summarize videos that are many hours long, be a coding teacher, do market research, research, research, and a ton more. And I bet that just one or two more papers down the line and it will be way faster and more reliable. And for me, this paper refers to dozens of previous works so I can have a glorious time going through them and learning a ton of thanks ChatGPT. I can't wait. What a time to be alive. 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 now get on demand H100 instances for just $2.40 per hour. They are one of the first cloud providers to offer publicly available on demand H100 access. 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.