OpenAI Releases ChatGPT Pro, New o1 Model and More!

Okay, this is a Great week for AI. We got updates across the board from OpenAI pushing the envelope with a brand new O1 Pro model under a $200 subscription. All the way to a teaser of world generation models from Google's DeepMind, a brand new top tier AI video generator that's open source. And so many more unique stories. In this week's episode of AI News, you can use the show that collects all of the AI stories that you can actually put to work today. We find them, research them, test them, and package everything for you in one easily digestible video. No time to waste, get into it. Okay, so the biggest story and piece of news that you can actually use this week is definitely OpenAI coming out with 01, their full version of their reasoning model. And they also released a brand new subscription plan called ChatGPT Pro, coming in at 10 times the price of the plus plan, $200 a month. I already upgraded to this plan to show it off to you in this video. Quick side note, I had to do it on a different OpenAI account as my main one because my main one is in a team subscription and I couldn't just upgrade my license within the team. I had to use a separate account. And when you do that, it looks very similarly to what it looked like just yesterday. You have the model switcher up here with two slight changes. One of them is that O1 preview changed to 01. This is the full version of 01 that we've been waiting for that performs way better on all sorts of benchmarks that you really care about. If you're using 01 and then you have O1 Pro, which is a marginal improvement in performance. But the consistency is the main selling point here, and you run a prompt ten times way more often, are you going to get consistently high quality results? Okay, so before we talk about the new 01 and what my first impression is of how these models perform, let me just point out the other things that come with this $200 a month subscription. Now you get unlimited access to all of the normal models, plus unlimited access to Advanced Voice mode. This is something that many people have been requesting. And although they did change this over the past week, some users even reported up to 90 minutes of usage sometimes. And even the window in which it resets got shrunk from 24 hours to somewhere around 16 to 20 hours, it always varied. Nevertheless, if you wanted to use Advanced voice mode as a passive assistant while working, you couldn't really do that because after about an hour it would just turn off. This is the first plan where you can really do that. You could have the advanced Voice Assistant on 24. 7, which I guess opens up a new set of use cases that nobody has really considered yet. If they pop up, we'll certainly cover them. But it is interesting that you could now have a 24. 7 voice assistant assistant on your phone. I don't think most people need that, just like most people don't need unlimited access to 01. Even if you hit the rate limit sometime, you just wait a little bit and you can run more prompts. I think the one big selling point right here is the access to the Pro mode. And as pointed out by costing our community here today, interestingly a thing they did not mention is that this Pro subscription comes with a larger context window. This is the first time you're getting 128,000 tokens of context inside of ChatGPT, which means you can effectively work with four times more text than you could about until now. That includes everything from input to output and the entire conversation as it goes on. And that's pretty much it. So for most people this is not going to be interesting. I'm going to talk about the models now and then I'll round out this segment with a recommendation of who should even be considering this obviously very expensive plan. So let's talk about the models here. And this is only a first look. This thing came out a few hours ago. I spent all of my time here running various test prompts through it and comparing it to the results of the previous models of GPT4,001 mini and then also making the comparison to 01 Full and 01 Pro. Now look, obviously this is just a first look and not a comprehensive testing, but I can already tell you that what you see here with these benchmarks like for example 01 and 01 Pro always being very close in the results right here. 76, 79, 78, 86 and these are competition math and PhD level science questions. Most people aren't asking those types of questions. Nevertheless, as we talked about, there's a lot of use cases that make a lot of sense inside of 01 even for everyday people. You might not be working with mathematics or scientific questions, or you might not even be using it for development, which by the way, it's really outstanding and probably even better than Sonnet. We will need some time to really state that confidently. But as I mentioned in another video on the channel where I took a deep dive into 01 and looked at what makes sense in for example a business and marketing context. It's really good at long term planning whenever you need to keep the bigger picture in mind and not just go step by step but really think about the outcome and the resources you have and bringing all of that together every step along the way. This is where these reasoning models are fantastic and where way superior to anything you'll ever get from a sonnet or a GPT4O. And as I also pointed out, O1 Mini is fantastic at rewriting. I didn't really see a difference in performance when it comes to rewriting, so we can just leave it at that for now. But I ran this fairly advanced prompt on how to create a SWOT analysis. It doesn't include examples in the end, but there is quite a bit of context in here that allows it to do a great job at this and I compared this between GPT 4.0 and even 01 Pro and let me tell you, my opinion didn't really change compared to what it looked like on a week ago. When it comes to these models, 01 preview is way better at actually thinking this through and as I said, considering various viewpoints and things like an outcome or division, even more than 4.0 will ever be able to do. And this didn't really change with 01. Matter of fact, I don't even see a big difference in this particular use case. I mean, fair enough, they advertise this as hey, this is really good for science, it's good for math, it's good for code generation. And I guess that's where all the gains were will be found. But just like before, I always went to O1 preview for a lot of these strategic type prompts. 01 seems to be doing equally as excellent on those tasks. And if I Compare this to 01 Pro mode, it's virtually the same. Except that the Pro model took even more time. I then did the same thing for Social media content calendar, which Zero1 Pro took over 90 seconds for, and Zero1 was faster. And again it essentially came up with the same thing and both of them are head and shoulders above what you can get from ChatGPT 4.0. So I don't know, I suppose the verdict is 01 is head and shoulders above chat ChatGPT 4.0 when it comes to some of these strategic use cases that I personally really care about. But the Pro mode not really necessary for most users. And again, that's just speaking from the limited tests I ran here. I mean, hey, I literally only ran one coding prompt, which is where I asked it to build a browser chess game, and it did that. And the Pro mode was slightly longer. I ran it through Tokenizer and the output was around 10% longer. But when I ran the same thing through Claude, the output was actually 30% longer. Plus obviously in the artifacts you could actually run it in the browser. But I realized that this doesn't mean at all. Just time will show how this actually performs in practice. This is just a first look, but nevertheless, my verdict for now would be I don't think Most people need 01 Pro. Matter of fact, if you look at the graphs, the main thing it has going for it is more reliability. That's where the real difference lies between 01 normal and 01 pro. If you run a prompt four times, 01 pro is going to be more reliable. So who's this $200 a month pro plan for? Well, over the years of learning all different types of software and buying all sorts of electronic hardware, I mean, I have three cameras just laying around on the table right now randomly, and that's because I'm selling some of the stuff that I don't need. But the point is this. I came up with a rule of thumb of my own when it comes to all of these technological upgrades, because there's always something you can upgrade. There's the new Pro model, there's the new iPhone model, there's the more expensive lens, whatever it might be. If you're already using a product and you find yourself running into the limitations three or more times in under a month, that and only that is a scenario in which you should even consider upgrading to a higher tier subscription or a more expensive version of the thing than that removes that restriction. In other words, if over the course of the last month you haven't hit the rate limits of O1 preview a single time, or maybe you only hit the rate limit of Advanced Voice twice, don't bother. Save your $180. It's not going to be a game changer for you. On the other hand, if you're using it for your work every day, or maybe used O1 preview for code generation every single day, and you were consistently annoyed by the fact that the limits came up and you can afford this more expensive subscription plan, it might benefit you. But I believe that for most people that did not hit these rate limits more than three times over the past month, this plan is the easiest skip. Although they might include more into it over the course of the next 11 days with these 12 days of OpenAI Christmas that they're doing okay so that's the news from OpenAI, but let's have a look at other stories because so much more interesting stuff that you can put to work today came out this week. Some of it arguably way more practical than these new 01 models for most people. Okay, next up we have something that came out last week, but I just want to point this out briefly. It's innovative thing that came out of Anthropic's claude and it's called the MCP protocol, AKA the Model Context protocol. Now this might sound fancy, but at the end of the day, this is just a way to link up external tools to your CLAUDE interface. You can think of this like the tools inside of ChatGPT, like here. It has the ability to search the web and as you might know, under the hood it has the ability to browse the web, run the code interpreter or generate images. Now what CLAUDE did with the Model Context protocol is actually it gave users the ability to implement custom tools. So if you want CLAUDE to build browse the Internet, which it could not do previously, you can hook it up via Model Context Protocol to an Internet browsing API. And now CLAUDE can do that. If you want to hook it up to your server or a weather app that pulls live weather data, you can also do that with mcp. There's a lot of possibilities here and I'm planning to actually spend my weekend exploring this as I definitely want to cover this more in between the channel and the community. It seems like a fantastic advanced way to get the most out of these.