OpenAI o3-mini - Thinking AI for Free…For Everyone!

OpenAI’s o3 mini AI is here, this  is their reasoning and thinking AI,   and I waited a little before publishing this  video to see how you Fellow Scholars are using it,   which is not very good for business and views, but  it’s better information to you Fellow Scholars,   and that’s what matters so we do that, and  I have to say, the results are incredible. The classic bouncing ball experiment  that could not be reliably done with   the previous OpenAI o1 system could be  done with the free DeepSeek R1 that you   can run at home self-hosted, so  what is OpenAI’s answer to that? Let’s see, oh yes, this one  can finally do that, but now,   let’s dream bigger. Let’s graduate to  10 balls. Still works. Fantastic, so,   can we do it? Can we ask for even a 100 balls?   Let’s see…it works beautifully, loving it. It can also create better, more interesting  Minecraft worlds compared to the previous version,   there is quite a difference, and if you want  to specify an input image and write a program   that creates this beautiful ASCII character art  from it, you can do that too in just one prompt. And get this, it is now available for free for all  of us, you can hopefully also try it by clicking   on the “Reason button” when it appears. If you  are interested, you should probably try it right   now. But the good news continue, so, Károly,  I hear you asking…do we have a paper? Oh yes,   fantastic, and it has some really surprising  results there too, more on that in a moment. I love looking at these comparisons, for  instance, this falling letter experiment   shows that other systems have also caught  up and are performing really well too. So, is o3 mini any good? And the answer  is yes. When looking at benchmarks,   things are going up and to the right in an almost  unprecedented manner. I mean, just look at that.    But it gets better. Paper time! Wow. I don’t  even know which the crazier news is. The paper   reveals that one, that even the older GPT-4o is  better than the baseline expert humans in biology   experimentation, or that even the mini version of  o3 outperforms the previous flagship thinking AI,   and it is so cheap to run that it is now  free to all users, with some limits. Yes,   all of us. Super cool. Note that we have no  business relationship with OpenAI. Never had. And there is an even bigger and more  surprising insight in this story. For many years we thought that whichever  company gets first to something that kind   of looks like a generally intelligent AI  system is going to be unbeatable. Today,   it does not seem to be so. Completely open source  solutions that you can run yourself at home for   free are popping up day by day, and for the first  time ever, they are able to challenge OpenAI. That   is absolutely amazing for us. Just think about  it: o1 took just a few months to reproduce,   and that thinking AI was a paradigm shift. But  o3 is not a paradigm shift, is it a better o1,   so I think it will be challenged by a  completely free variant in less than 3 months,   possibly even less than a month. So how  is all this possible? Well, and all this   is possible through the power of the papers and  open sourcing and open science. People are able   to work together all around the world,  share their findings with each other,   and they put together all their smarts to create  something even better. That is humanity at its   best. And, as a result, we are all going to have  intelligence in our pockets for free. Loving it. That is what we have been talking about for almost  10 years and nearly a thousand videos, and now,   suddenly, everyone is talking about the papers.   What a huge win! That makes me very, very happy. So that is the amazing world we live  in today. What a time to be alive! Also note that in the system card, areas  of risk are reported, and it’s not just   sunshine and rainbows, I like how these  are reported in such a transparent manner,   respect for that. Also, if OpenAI’s o1 was a  formidable con artist, I don’t know what o3   is because it is so much stronger in that area.   I hope that this can and will also be used as a   shield against this seemingly neverending  line of people who have coins and riches   to offer to anyone who believes them. Using AI  as a shield. Somebody please work on that too. So, conclusions. AI research keeps progressing  at an absolutely stunning pace, these systems

are getting more and more intelligent, they will  be able to help you build a billion dollar company   completely alone, and all this is going ton run  in your pocket, for free. I mean, just while I   started making this video, they also came up with  Deep Research, which improves the already improved   benchmark results on a proper dataset, part of  which is private, so hopefully it cannot be gamed   as easily. And it not only improved it, it doubled  it. My goodness. And, look: not a few weeks,   but a few hours later, an open source clone  has already appeared. What a time to be alive! Dear Fellow Scholars, this is Two Minute  Papers with Dr. Károly Zsolnai-Fehér. This   is where we celebrate amazing papers  and amazing human achievement together. Ah, and one more thing: every now and then we are  able to invite new sponsors to the show. This time   is that that. If you are interested, let us  know, the link is in the video description.