# OpenAI’s ChatGPT Took An IQ Test! ## Метаданные - **Канал:** Two Minute Papers - **YouTube:** https://www.youtube.com/watch?v=2A9PLW6BCx4 - **Дата:** 18.01.2023 - **Длительность:** 7:12 - **Просмотры:** 353,711 - **Источник:** https://ekstraktznaniy.ru/video/13325 ## Описание ❤️ Check out Weights & Biases and sign up for a free demo here: https://wandb.com/papers Try ChatGPT here: https://openai.com/blog/chatgpt/ My latest paper on simulations that look almost like reality is available for free here: https://rdcu.be/cWPfD Or this is the orig. Nature Physics link with clickable citations: https://www.nature.com/articles/s41567-022-01788-5 SVG objects: https://twitter.com/brdskggs/status/1599533975357095936 Twitter logo with SVG graphics (original): https://twitter.com/faridsabitov/status/1605612737886687248 Twitter logo with SVG graphics (not the original): https://twitter.com/DataChaz/status/1605989977447272448 Shadertoy: https://www.shadertoy.com/results?query=chatgpt Mirror shader: https://sigmoid.social/@FidIsTheOne@mastodon.gamedev.place/109558911131170252 Security holes: https://twitter.com/moyix/status/1598081204846489600 Failing at multiplication: https://twitter.com/Dhilung/status/1598168648392519681 IQ test: https://davidrozado.substack.com/ ## Транскрипт ### Intro [] Dear Fellow Scholars, this is Two Minute  Papers with Dr. Károly Zsolnai-Fehér. This is the amazing ChatGPT, a  variant of the GPT-3 language   model AI that was unleashed  to read the internet and now, ### ChatGPT [0:14] we can use it as a conversational agent. And  today, we can even subject it to an IQ test. In a previous episode, we saw that  it could write new song lyrics,   or even give us ideas for designing a living  room. However, you see, this is a language model,   so it can only give us text. Not images. However,  no one said that we cannot plug in these text   prompts into a text to image AI, so these Fellow  Scholars did just that. And thus, we could even   see these beautiful results come to life. This  was one of my favorite examples, and it turns out,   we can even get ChatGPT to give us some beautiful  imagery with some other tricks. And it can do ### Examples [1:06] even more. Here are 5 more amazing examples that I  absolutely adore and would love to share with you. One, we can ask it to create the description of  an SVG file, which is a vector image format. This   describes in text what kinds of objects go where.   And, are you thinking what I am thinking? Oh yes,   this is an excellent way of using text to  create images. So it can create this red circle,   blue triangle, or even a face. Now if you  think that it is all it can do, then have a   look at this. For instance, can it create a vector  version of the Twitter logo? Wow, it can. So cool! Okay, but I am a computer  graphics researcher by trade,   so I would love to see something a little more  advanced. So what other piece of text is there   that can create absolutely beautiful images? Of  course, computer code! And hold on to your papers,   because here it is. Two, it can write proper  pixel shaders. What is that? This is a website   called shadertoy where we can submit a piece of  computer code that is executed by our graphics   cards and produces an image. Oh yes, an even  more powerful way of taking a piece of text,   and creating beautiful images, videos, or  even a computer game with it. And ChatGPT   can even write these cool little programs  too. They compile and run well. Wow. Now, you may be thinking, are these  just some cute little test programs,   or are they actually useful in any way? Well,  three, check this out: this example shows that it   can even write computer code to place a mirror in  our computer game. It works like this. Yet another   example of using a tool for something that it  was not designed to do. I absolutely love these. Four, when we are talking about potential  applications in the industry, here is another one.    Whoa. ChatGPT can be asked to find, and even fix  security holes in a piece of computer code. That   is such an incredible thing because it requires  knowledge on what happens if we execute a piece of   code, and even better, what potential weaknesses  may arise. That feels like talking to an   intelligent being…until we ask it to multiply two  numbers together, and it then fails spectacularly.    It is so close! But not quite correct. The amusing  properties of a modern AI system. Loving it. Five, so actually, how reliable is it? Well,  it is not always that unreliable. A Fellow   Scholar had ChatGPT take the SAT test, and  this, dear Fellow Scholars is just about   the SAT score for an average human. Wow.   And it can also take a linguistic IQ test, ### Conclusion [4:24] and if you have a closer look, these questions  require a considerable English vocabulary. So,   what does this test think ChatGPT’s IQ  is? It says 147. Holy mother of papers.    Now before we get too excited, note that  these are not peer-reviewed experiments,   so I am a little hesitant to give them too  much credence, and I hope you are too. So   please don’t take these too seriously. However,  I feel that these are likely early indications   of something amazing coming up real soon. I  really hope that we will see a few peer-reviewed   papers on the academic capabilities of  ChatGPT in more rigorous tests soon. So the original GPT was an AI that was meant to  read the internet, learn English, and maybe be   able to finish incomplete Amazon reviews. And  now it knows computer graphics programming,   cybersecurity, has lots of  general human knowledge,   and more. Just imagine what this will  be able to do a couple more papers down   the line. If you have some ideas,  let me know in the comments below! And now, if you have a Mac computer, you can even  add it to your menu bar to access it quickly,   and for free. How cool is that? Or, plugins for  all major browsers now seem to be available too.    So, until the follow up papers appear, let  the experiments begin! ChatGPT has inspired   a lot of you Fellow Scholars all over the  world, and has reached a million users in   an incredibly small amount of time. That was  one of the most explosive transitions from a   paper to a real-world product to show everyone  that the papers are real. As real as it gets. Thanks for watching and for your generous  support, and I'll see you next time!