OpenAI’s New Free AI: The Good, The Bad, The Unexpected!

This is Open AI's open AI model, and it's a bit like finding a fully functional space shuttle in your garage next to the lawn mower. Yes, this is history in the making. Finally, OpenAI published an openweight AI model, GPT OSS. Of course, it passes the classic vibe check, the bouncing ball, and my little forest fire simulator with little dots as animals is also working really well. This is a resilient ecosystem. I have to say the fires are barely spreading. Everyone is talking about this AI. There are two versions, one for high-end laptops and one medium-sized version that can run on lower-end computers. Wait one more week and I almost guarantee that you fellow scholars will run it on your phones in your pocket and it performs superbly on a variety of tests, including my favorite, humanity's last exam. This is the toughest test for the toughest AIs out there. Questions from smart scientists from all around the world that no AI could answer at the time it was made. The closed GPT40 model was able to get less than three questions out of a 100, right? And now we get something for free that can solve 19 of them. Just think about that for a moment. That is insanity. And now here are three things that very few people talk about and are super important. One, oh my goodness, look at that. For me, it performs unexpectedly well on health-based questions. Look, this is not normal. It rivals one of the best paid proprietary solution 03 in these cases. That is unbelievable. And of course, it is not infallible, but imagine that you can have this on your laptop and take it anywhere in the world with you if there are any emergencies. And I don't have any connectivity. This can be a little personal doctor. I am sure it will save many lives. This is a huge gift and a great contribution to humanity. Thank you so much. Two. Uh-oh. Hallucinations. In these tests, they make up stuff in the majority of their answers. Now, note that these particular tests are asking for lots of niche information, and these smaller world models just don't have this kind of knowledge within them. But still important to note, it is super simple to run. There is a free version of it. The link is in the description. As you see, I ran some very scholarly experiments with it already. or to run it privately. I fired up a quick lambda instance on a rented GPU. But note that this has no multimodel support. You can't give it images to look at text only. And here is what I found to be the most unexpected of them all. Dear fellow scholars, this is two minute papers with Dr. Carola or Dr. Carol John Alpha, a version that you can pronounce and write easily. And by the way, around here, we don't lean into drama or waste your time with stretched out long videos for more revenue. So, I would like to kindly ask you fellow scholars to watch this video until the end and leave a kind comment. That is the only way to get YouTube to recommend these videos. I found this yesterday from one of you fellow scholars and it really warmed my heart. So three, initially I thought it surely costed billions of dollars to train this right now. Hold on to your papers, fellow scholars, and let's see. Would you look at that? Wow, that is hugely unexpected. Based on this information, I would roughly estimate that it costs less than $10 million to train the bigger model and maybe even less than a million for the smaller one. But that means something even more amazing. It means that we are likely going to see many more free AI models popping up with similar capabilities. There is going to be huge competition in this area. And I think just a few months from now, we are going to have a huge ecosystem of equivalent or perhaps even better open models. Super excited for that. But that is just one direction. It is also customizable. So even in the shorter term, you smart fellow scholars are going to fine-tune these models for specific applications like legal contract analyzers, biotech literature miners, academic peerreview assistants, and so much more. What a time to be alive. And I am thinking about making a follow-up video about the key technical innovations in GPT OSS. Make sure to subscribe and hit the bell to get notified. And I got to say, I can't stop playing with OpenAI's Open GPT model through Lambda GPU cloud. And as you see, I am doing very useful things with

it for science. Yes, this is actual speed. I can't believe that I can have more than 100 billion parameters running super fast here. Many of you fellow scholars are using it and if you don't, make sure to check it out. It costs only a couple dollars per hour. Insanity. You can rent an Nvidia GPU through lambda. ai/papers