The first day of the 12 Days of OpenAI is here, and we've got our first announcements, and it is huge! 001 is ready. Let's watch the video; I'll react to it. Let's get into it. All right, here we go. We have some members of the OpenAI team.

Really, I think the only one I recognize is obviously Sam Altman. Let's start watching. Hello! Welcome to the 12 Days of OpenAI. We're going to try something that, as far as we know, no tech company has done before, which is: every day for the

next 12—every weekday—we are going to launch or demo some new thing that we built, and we think we've got some great stuff for you, starting today. We hope you'll really love it. Let's pause for a second. Right away, look at that sticker!

The strawberry! I love it! What a good sign for things to come for these next 12 days. You know, we'll try to make this fun and fast and not take too long, but it'll be a way to show you what we've been working on—a little holiday

present from us. So we'll jump right into this first day. Uh, today we actually have two things to launch. The first one is the full version of 001. We have been very hard at work. We've listened to your feedback. You wanted—you liked the OpenAI

preview, but you want it to be smarter and faster and be multimodal and be better at instruction following—a bunch of other things. So we've put a lot of work into this, and for scientists, engineers, coders, we think they will really love this new model. Uh, I'd like to show you

Quickly, about how it performs. All right, so there it is. There's the big news: we have the full version of 01 ready to go. And of course, yeah, you all want it faster, better, and smarter. And yeah, of course, we all want these things, but...

Uh, it's funny how he kind of lists that off like it's no big deal. Uh, but no, they're delivering. So let's take a look. So they're about to show us these different benchmarks from the 01, the 01 preview, and GPT-4 against three major

high-end benchmarks: this is math, coding, and PhD-level science. So let's hear what he has to say. So you can see the jump from GPT-4 to 01 preview across math, competition coding, and GP QA Diamond. Um, and you can see that 01 is a pretty big step

forward. Um, it's also much better in a lot of other ways, but raw intelligence is something that we care about. Coding performance, in particular, is an area where people are using the model a lot. So in just a minute, uh, these guys will demo some things about A1. They'll show

you how it does at speed, how it does at really hard problems, how it does with multimodality. But first, I want to talk just for a minute about the second thing we're launching. All right, before I let him get to the second thing that they're

launching... yeah, I mean, look at this massive jump in math from the AIM 2024 competition math benchmark. We have 13.4 in GPT-4, all the way up to 56.7 in 01 preview. And another massive jump...

up to 83.3 and 01, so I honestly wasn't expecting 01 to be such a big leap from the 01 preview, but it really is. Now, the only benchmark where it didn't and actually went down is from the 01 preview to 01 on

the PhD-level science questions; it actually went down a little bit. In fact, this is the least big jump from GPT-4 to the 01 series of models, so it's pretty interesting. However, an expert human still does not score as well as these 01 models, so pretty interesting.

stuff. It looks like we're getting just another massive leap in thinking models. Let's keep watching. A lot of people—uh, power users of ChatGPT at this point, they really use it a lot—and they want more compute than $20 a month can buy, so we're launching a new tier, Chat

GPT Pro. Pro has unlimited access to our models, uh, and also things like advanced voice mode. It also has a, uh, a new thing called 01 PR mode. So 01 is the smartest model in the world now, except for 01 being used in PR mode, and for the

hardest problems that people have, 01 PR mode lets you do even a little bit better. Um, so you can see at competition math, you can see at GSM QA Diamond, um, and these boosts may look small, but in complex workflows where you're really

pushing the limits of these models, it's pretty significant. All right, so let me pause for a second, and I think I understand what's going on. So first, they're launching a new subscription tier, and we've heard rumors of this from a few months back. I reported on it and

It looks like it has come to fruition. They have a $20/month plan, and now they have a $200/month plan; that is a 10-times increase in the price. But what you get for that is unlimited everything. And not only that, they are essentially

giving you more compute during test time—compute to get these increased scores in the O(1) PR mode, as what we're seeing here. So that's basically, from my understanding, how it's working: they're essentially allowing the model to think more; they're giving you more compute, and

of course, they're charging you a lot more, so they should give you more compute. But it's super impressive and actually speaks volumes towards how scalable test-time compute really is and how we now have this new lever of scalability on the post-training side of

a model, and I love it. That's super exciting for me. So let's keep watching. One thing that people really have said they want is reliability, and here you can see how the reliability of an answer from Promp mode compares to O(1), and this is

an even stronger delta. And again, for our Pro users, we've heard a lot about how much people want this, so I find this really interesting. And this is probably something I need to do a little bit more research on myself, which is: when you're running one of these models against a

benchmark and you get a score, I feel like the reliability should be baked into the benchmark itself, not having to run the same question over and over again to see what the reliability is. I always thought the reliability was built into the benchmark, which is odd, and so I

Need to do a little more digging into how these benchmarks actually operate, but what we're seeing here is that when you run it more and more times, the reliability is much higher on the 01 Pro mode. So that extra compute during test time actually does make a difference for

reliability. Chat GPT Pro is $200 a month. Uh, it launches today. Over the course of these 12 days, we have some other things to add to it that we think you'll also really love. Um, but unlimited model use and this new 01 Pro mode, so

should I spend $200 a month for 01 Pro mode? Are you going to let me know in the comments? I want to jump right in, and we'll show some of those demos that we talked about. Uh, and these are some of the guys that helped build 01, uh, with

many other people behind them on the team. Thanks, Sam. Hi, um, I'm Hungan. I'm Jason, and I'm Max. We're research scientists who worked on building 01. 01 is being rolled out today to all Plus and soon-to-be Pro subscribers on Chat

GPT, replacing the 01 preview. The 01 model is faster and smarter than the 01 preview model, which we launched in September. After the launch, many people asked about multimodal input, so we added that. Uh, so now the 01 model, live today, is able to

reason through both images and text jointly. Okay, they kind of just put that under the radar, almost like it wasn't a big deal. Now 01 is multimodal; that is one of the biggest upgrades that everybody has been waiting for. So really cool to see. As Sam mentioned, today we're

Also, going to launch a new tier of ChatGPT called ChatGPT Pro. ChatGPT Pro offers unlimited access to our best models like GPT-4, and advanced voice. ChatGPT Pro also has a special way of using GPT-4 called GPT-4 Pro mode. With GPT-4 Pro mode,

you can ask the model to use even more compute to think even harder on some of the most difficult problems. Yeah, so that that's exactly what I thought it was going to be. I'm glad he confirmed it. Essentially, you're just saying, "Give me more compute." I wonder what the

limits of that are. Is it limited? I wonder how much they've tested. What if you let it think for a day, a week, a month? Very interesting to think about. We think the audience for ChatGPT Pro will be the power users of ChatGPT, those who are

already pushing the models to the limits of their capabilities on tasks like math, programming, and writing. It's been amazing to see how much people are pushing GPT-4 preview—how much people who do technical work all day get out of this, and uh, we're really excited to let

them push it further. Yeah, sure. We also really think that GPT-4 will be much better for everyday use cases, not necessarily just really hard math and programming problems. In particular, one piece of feedback we received about GPT-4 preview constantly was that it was way too slow.

It would think for 10 seconds if you said "hi" to it. And we fixed that. That was really annoying. It it was kind of funny, honestly. It really thought it cared; really thought hard about saying hi back. Yeah. Um, and so we fixed that. GPT-4 will now

think much more intelligently. If you ask it a simple question, it'll respond really quickly, and if you ask it a really hard question, it'll think for a really long time. Uh, we ran a pretty detailed suite of human evaluations for this model, and what we found was that it

made major mistakes about 34% less often than GPT-3.5 preview, while thinking fully about 50% faster. And we think this will be a really, really noticeable difference for all of you. So I really enjoy just talking to these models. I'm a big

history buff, and I'll show you a really quick demo of, for example, a sort of question that I might ask one of these models. All right, so that's a really cool update. And again, I don't use GPT-3.5 for many use cases, only the most complex use

cases where I'm really trying to think through hard problems—math problems, logic problems—do I go to GPT-3.5? But if they really make it faster and more efficient to use and smarter in terms of how much thinking it puts towards some of these easier problems, then, of course, like I'm

going to go to GPT-3.5 more often. And then now thinking about it, that $200 a month price point maybe it makes sense. So let's keep watching. So, so, uh, right here, I, on the left, I have GPT-3.5; on the right, I have GPT-3.5 preview. And I'm just asking it a

really simple history question: list the Roman Emperors of the second century; tell me about their dates, what they did. Um, not hard, but, you know, GPT-4 actually gets this wrong a reasonable fraction of the time. That's surprising to hear; this seems like something GPT-4 would get

Right, instantly and often. And so I've asked—one, this I've asked—preview this. I tested this offline a few times and I found that 01, on average, responded about 60% faster than 1 preview. Um, this could be a little bit variable because right now

we're in the process of swapping all our GPUs from 01 uh Pro preview to 01. So actually, 0 thought for about 14 seconds; 01 preview still going. There's a lot of Roman emper—there's a lot of Roman emperors.

Foro actually gets this wrong a lot of the time. There are a lot of folks who rolled for, like, uh, 6 days, 12 days a month, and it sometimes forgets those. Can you do them all from memory, including the six-day people? No. Yep. So, so here we go. 01 thought for

about 14 seconds; 01 preview thought for about 33 seconds. These should both be faster once we finish deploying, but we wanted this to go live right now. Exactly. Um, so yeah, we we think you'll really enjoy talking to this model. We found

that it gave great responses; it thought much faster; it should just be a much better user experience for everyone. So, one other feature we know that people really wanted for everyday use cases that we've had requested a lot is multimodal inputs and image understanding, and Hungun is going to

talk about that now. This is one that I've been really looking forward to. Without having multimodal understanding, the 01 model can only go so far for my personal use cases. So let's keep watching. I created this toy problem uh with some hand-drawn diagrams and so on.

So here it is. It's hard to see, so I already took a photo of this, and let's look at this photo on a laptop. So, once you upload the image into the chat GPT, you can click on it and see the zoomed-in version. So this is a

system of a data center in space. So maybe, um, in the future we might want to train AI models in space. Uh, I think we should do that, but the power number looks a little low. One gigawatt? Okay. But the general idea... rookie numbers. This rookie

numbers, rookie. Okay. Yeah. So, uh, we have a sun right here, uh, taking in power on this solar panel, and then, uh, there's a small data center here. It's exactly what they look like. Yeah, GPU, and then pump—nice pump here. And one interesting thing

about GPU operation in space is that on Earth we can do air cooling, water cooling, to cool down the GPUs, but in space there's nothing there, so we have to radiate this heat into deep space. And that's why we need this giant radiator cooling panel. And this problem is about finding the lower bound estimate of the cooling panel area required to operate this one-gigawatt data center. Probably going to be very big. Yep. Let's see how big. Let's see. So that's the problem I'm going to

prompt, and, uh, yeah, this is essentially asking for that. So this is awesome! This is so cool. Just a hand-drawn, very complex problem; not much information in the actual schematic, but then they ask very specific questions over text. And let's see how it does. Let me hit go, and the

prompt and uh yeah this is essentially asking for that so this is awesome this is so cool just a handdrawn very complex problem not much information in the actual schematic but then they ask very specific questions over text and let's see how it does let me uh hit go and the

The model will think for a few seconds. By the way, most people don't know I've been working with Hangan for a long time. Hangan actually has a PhD in thermodynamics, which is totally unrelated to AI, and you always joke that you haven't been able to use your PhD

work in your job until today. So you can trust Hangan on this analysis. Finally, thanks for hyping it up; now I really have to get this right. Uh, okay, so the model finished thinking in only 10 seconds. It's a simple problem, so let's

see how the model did it. So, power input... So first of all, this one-gigawatt figure was only drawn from the paper, so the model was able to pick that up nicely. And then, radiative heat transfer only—that's the thing I mentioned—so in space,

nothing else. And then some simplifying choices. And one critical thing is that I intentionally made this problem underspecified, meaning that the critical parameter—the temperature of the cooling panel—I left it out so that we can test the model's

ability to handle ambiguity and so on. So the model was able to recognize that this is actually an unspecified but important parameter, and it actually picked the right range of temperatures, which is about room temperature. With that, it continues the analysis and does a whole bunch

of things. And then, wow, yeah, look at this!  A simple problem... oh my goodness, this is amazing! It found the area, which is 2.42 million square meters. Just to get a sense of how big this is, this is about 2% of the land area of San Francisco.

This is huge, not that bad, not that bad. Yeah, okay. Um, yeah, so I guess this...this uh...reasonable. I'll skip through the rest of the details, but I think the model did a great job, um, making nice, consistent assumptions that, um, you know, make the required area as little as

possible. And so, um, yeah, so this is the demonstration of the multimodal reasoning, and this is a simple problem, but O1 is actually very strong, and on standard benchmarks like MMLU and MathVista, O1 actually has state-of-the-art

performance. Now Jason will showcase the...the pro mode. Great. So I want to give a short demo of CHBT O1 pro mode. Um, people will find O1 pro mode the most useful for, say, hard math, science, or programming problems. So here I have a pretty

challenging chemistry problem that O1 preview gets usually incorrect. And so I will uh...let the model start thinking. Um, one thing we've learned with these models is that, uh, for these very challenging problems, the model can think for up to a few minutes. I think for this

problem, the model usually thinks anywhere from one minute to up to three minutes. Um, and so we have to provide some entertainment for...for people while the model is thinking. So I'll describe the problem a little bit, and then if the model's still thinking when I'm done,

I've prepared a dad joke for...for us, uh...to fill the rest of the time. Um, so I hope it...for a long time. You can see, uh, the problem asks for a protein that fits a very specific set of criteria. So there are six criteria, and the challenge is...So one cool thing...

While he's explaining, I'll just pause for a second. I haven't seen this UI element yet, where it's kind of a progress bar, and so it has a predetermined, I guess, how long it's going to take, and it's going through it. Maybe I've missed that, but I haven't personally seen that progress bar UI element yet. Let me know if you have.

Each of them asked for pretty chemistry domain-specific knowledge that the model would have to recall. And the other thing to know about this problem is that none of these criteria actually give away what the correct answer is. So, for any given criteria, there could be dozens of proteins that might fit that criteria, and so the model has to think through all the candidates and then check if they fit all the criteria.

So, is this a dig at Google and their AlphaFold project? Is this saying, like, 'Hey, we can do that now with the O1 model'? I'm not sure; this is kind of out of my area of expertise, but yeah, let me know.

Okay, so you could see the model actually was faster this time. Uh, so it finished in 53 seconds. You can click and see some of the thought process that the model went through to get the answer. Uh, you could see it's thinking about different candidates, like neuroligin initially, um, and then it arrives at the correct answer, which is retinoic acid.

Uh, which is great. Um, okay, so to summarize: we saw from Max that O1 is smarter and faster than O1 preview; we saw from Hangan that O1 can now reason over both text and images; and then finally, we saw...

With ChatGPT Pro mode, uh, you can use GPT-4 to think about, uh, the, the, to, to, to reason about the hardest, uh, science and math problems. Yep, there's more to come, um, for the ChatGPT Pro tier. Uh, we're working on even more computer-intensive

tasks to power longer and bigger tasks for those who want to push the model even further. And we're still working on adding tools to the GPT-4 model, such as web browsing, file uploads, and things like that. That's going to be huge; that is the other piece I needed: multi-modal

and I also needed to have tools. Just web browsing alone would be a huge, huge gain, so very excited about that. We're also hard at work to bring GPT-4 to the API. We're going to be adding some new features for developers: structured outputs, function calling, developer

messages, and, uh, API image understanding, which we think you'll really enjoy. We expect this to be a great model for developers and really unlock a whole new frontier of agentive things you guys can build. We hope you love it as much as we do. That was great. Thank you guys so much.

Congratulations, uh, to you and the team on, on getting this done. Uh, we, we really hope that you'll enjoy GPT-4 and Pro mode, uh, or Pro tier. Uh, we have a lot more stuff to come. Tomorrow we'll be back with something great for developers, uh, and

and we'll keep going from there. Before we wrap up, can, can we hear your joke? Yes. Uh, so, um, I made this joke this morning. The, uh, the joke is this: so Santa was trying to get his large language model to do a math problem, and he was

prompting it really hard, but it wasn't working. How did he eventually fix it? No idea. He used reinforcement learning. [Laughter] Thank you very much. Thank you. All right. Yeah, that was a dad joke, to say the least. So what do you think is

going to come over the next 11 days? A few things that I personally would be very excited about: any kind of preview of their browser, or kind of more of an agentive system that can actually browse the web. I would love to see it. I know Operator has been rumored to be coming

soon, so I really want to see something from OpenAI where agents can actually take action on your behalf. I'd love to see an update to Dolly; that would be amazing. Obviously, a Sora release would be amazing, but I don't really think that's going to be coming—maybe it will

given it was leaked like a week ago. But those are the things I'm most excited about. Let me know in the comments what you're most excited about; what do you want to see over the next 11 days? Should I keep covering all of these videos?

Let me know if you enjoyed this video. Please consider giving a like and subscribe, and I'll see you in the next one.

o1 PRO Mode - ChatGPT Pro with Unlimited Compute (Announcement Breakdown)

Artificial Intelligence (AI), Open Source, Generative Art, AI Art, Futurism, ChatGPT, Large Language Models (LLM), Machine Learning, Technology, Coding, Tutorials, AI News, and more

** Exclusive Pinecone Partner **

o1 PRO Mode - ChatGPT Pro with Unlimited Compute (Announcement Breakdown)

Scripts