OpenAI Finally Introduces NEW MODELS! (Updated GPT-4) + Leaked Updates

So OpenAI finally released some new things on their base model and that is of course GPT4. And this is not clickbait, it is not a leak, this is an actual release by the OpenAI team. So let's get into everything they did release. I mean it isn't some crazy crazy update, but it still is pretty important. So let's take a look at the new features. So well, first new feature that they do talk about was a new GPT 3.5. So they've said we've updated GPT 3.5 Turbo and and lower pricing. So if you don't know what GPT 3.5 is because you've been using GPT 4 so much, GPT 3.5 is the base free model that you can use when you go into chatgpt.com and of course this model is being upgraded. So they said next week we are introducing a new GPT 3.5 Turbo model GPT 3.5 Turbo 0125 and for the third time in the past year we will be decreasing prices on GPT 3.5 Turbo to to help our customers scale input. Prices for the new model are reduced by 50% to 0.0005 cents per 1000 tokens and output prices are reduced by 25% to 0.0015 cents per 1000 tokens. This model will also have various improvements including higher accuracy at responding in requested format and a fix for a bug which caused a text encoding issue for non English language functional calls customers using the pinned GPT 3.5 turbo model alias Alias will be automatically upgraded from the GPT 3.5 Turbo 0613 to GPT 3.5 Turbo 0125 Two weeks after this model launches. So you can see that the base model of ChatGPT 3.5 is going to be upgraded. Now one thing that would be fascinating is to see where this new version of GPT 3.5 does benchmark. They did say that it's going to be various improvements. So they've actually talked about the fact that these improvements aren't actually documented in this blog post because it does say various improvements and they said including higher accuracy at responding in requested format. So I think what they're going to do is they're going to try and bring up GPT 3.5 to be better than Google's Gemini Pro. Because of course, as you know, Google's Gemini Pro and Google's Gemini Ultra have just stolen the limelight. So I'm guessing that this upgraded GPT 3.5 will, which is still very, very good, is going to be an updated version that does surpass many other open Source and other LLMs on various benchmarks. So I wouldn't be surprised if that is the case. Now let's get to some of the juicy stuff and namely the point that many people are going to be looking at and that is of course the new GPT4 model. So they said that we are going to update GPT4 Turbo. So they essentially said that over 70% of the requests from the GPT4 API customers have transitioned to GPT4 Turbo since since its release, as developers take advantage of its updated knowledge cutoff, larger 128k context window, and of course the lower prices. Today we're releasing an updated GPT4 Turbo Preview model, GPT0125 Preview. This model completes tasks like code generation more thoroughly than the previous preview model and is intended to reduce cases of laziness where the model doesn't complete a task. So the thing that is the main focus here is of course the reducing of the laziness. So something that has been documented, and this wasn't just documented based on user experience, but was actually documented in a Stanford study. It actually did show that GPT4, the base version that was released in last year, was actually much later than the version that we do have now, which essentially doesn't complete tasks as well, and it doesn't do as long generations. Now, we don't know why this is, but many people are presuming that OpenAI may have changed the base model because of compute due to the expensiveness of the model. And that is of course a real thing. We know that the smarter a model is, the bigger it is, and the bigger it is, the more energy consumption it does take, the more compute it takes. Because of that, of course, we do know that if you've got millions and millions of people around the world, then of course we have something that isn't really profitable or possible to run. So of course, essentially what we need to do here is take a look at how this model has evolved, how much better it actually is, and if it is like the GPT4 base version that we did get and at launch, because I'm sure many of you who did actually interact with that model and did use the model know just how good it was. Now it will be also interesting to see if code generation does actually surpass on some new benchmarks, because of course, as we do know, a lot of people do like to do testings on before and after. So I'm guessing that a lot of people are going to be doing some tests on the GPT4 Turbo, the one that we have now compared to the new version, to see how much better it does on the code generation outputs. And it does also say here that for those who want to be automatically upgraded to the new GPT4 Turbo Preview versions, we are also introducing a new GPT4 Turbo Preview model name Alias, which will always point to our latest GPT4 Turbo Preview model. Now, what was absolutely crazy as well was that they said that we plan to launch GPT4 Turbo with vision in general availability in the coming months. So I'm guessing that the API for GPT4 Turbo with vision is going to be coming fairly soon. So yeah, overall what we've seen here is that the new Turbo model that was announced on Dev Day seems to have taken taken everyone in terms of the usage. So this is the model that everyone is using. And of course they're updating this model to essentially just be better than it was before. Now, I'm not sure if they're actually going to put this into the base chatgpt that everyone uses, because maybe this one might be more expensive. They didn't actually say that they were reducing any prices of this, so I'm guessing that perhaps you're going to have to just use GPT4 Turbo in either the playground or through the API. But maybe they might just roll this out altogether. But either way, this is good. And of course this is exactly what we expect when we know OpenAI are going to do release, because they always did talk about the fact that they were going to be doing some very gradual releases when it comes to their product launches. Now in addition, what they did actually talk about was of course the updated moderation model. So the free moderation API allows developers to identify potentially harmful text as part of our ongoing safety work re Releasing Text Moderation 007, our most robust moderation model to date. The text moderation latest and text moderation stable aliases have been updated to point to it and learn more about their safety practice guide. Now, there was also something to cover which is of course the API keys. So they said there are new ways to understand API usage and manage API keys. And we are launching two platform improvements to give developers both more visibility into their usage and control over API keys. So firstly, developers can now assign permissions to API keys from the API key page. So for example, a key could be assigned read only access to power, an internal tracking dashboard or restricted only access to certain endpoints. And secondly, the usage Dashboard and usage export function now exposes those metrics on an API key level after turning on tracking. So essentially you can view exactly how much that API key is using, and you can essentially see that per feature, per team, per product or project level, simply by having separate API keys for each. And if you want to know what that dashboard does look like, you can take a look over here, and this is what this dashboard does look like. So you can see the top organizations, the top API keys. You can see they've got a laundry buddy key, a playground key and untracked keys. And you can start tracking your keys by creating new usage, by creating new keys. And then of course, with this updated dashboard. So I think this is really, really good. I think this is really, really effective because we're now going to be seeing exactly where that usage is coming from. Now, of course, another thing that we need to do is we need to take a look at some of the future features. So something that wasn't exactly announced, but it is, is something that is likely going to be coming is of course the future features in terms of GPTs. Now, several people on several forums have been looking at what is in OpenAI's code and doing so, and in doing so they've been able to see some of the future features. And that's how we were able to identify the fact that OpenAI are working on a personalization and memory feature for ChatGPT. Now, essentially a new feature that they've talked about is of course the ability to mention a GPT and add it directly into your conversation. So I'm guessing what they want is they want us to be able to use multiple GPTs in different conversations. So here you can see that this person testing catalog, and I'll leave a link to the original tweet and a bunch of other stuff because there is so much information on this. So you can see right here that you can say, type to mention a GPT and directly add it into your conversation. Now, remember, currently this is a future feature, or it might just be a feature that is scrapped altogether because this is something that is rolled out experimentally, meaning that it's not really confirmed and it's just in a kind of testing phase. So essentially what we can see here as well is that when you do type at, you can see that we can search recent and pinned GPTs. And the reason that this is going to be, I guess you could say something that you want to have is because it's going to add to your workflow. So what you might want to do is you might want to use Filmmaker GPT to write a script within that chat. Then you want to add a different GPT in order to create an image from some things that you've been adding. And you can mix and match in certain different ways. And this screenshot right here. So, for example, ChatGPT now create a banner showing AI which achieved AGI in the form of a spear. And you can see robocam Filmmaker GPT is that one there. So, I mean, I think these future features that ChatGPT are going to be adding are going to be really, really cool. And I will leave a link to the article which essentially shows you how, you know, people were able to find these. We did actually talk about this in another video. I managed to find someone on Twitter who was just digging through the code that gets constantly updated and that's how we found the personalization. And I'm sure that's similarly how they found this feature here. Now, let me know how you're going to be combining these GPTs, because I think what I might do is I might just, you know, combine the knowledge of some GPTs, because I think that would be really effective, you know, for some, you know, one that are really specialized in marketing and some of the terms and guidelines, I think all having it in one context window and is going to be a lot better than having to switch from A and then switch to B. That is something that, of course, you know, it's just going to help your workflow even more. But let me know what you think about this. I think this update is really good in terms of receiving an actual big updated GPT4 model. I think GPT 4.5, which apparently may or may not be released very soon, will be released, if I had to guess, after Gemini Ultra does get released, depending on when that is. Some people are speculating that that might be released next week based on the fact that Google said it would be released in the last week of January, or at least near Q1. So I'm guessing that Google don't really have that much time left because of course it is a current race. So whilst Google is trying to roll out Jabili Ultra, we don't know when that is. I think that for now, GPT's Turbo's new model is going to have to give us some kind of entertainment in terms of seeing how much less lazy it is. Because you trust me, I actually recently did test GPT4 and how lazy it is and Trust me when I say it's incredibly lazy in terms of trying to do certain tasks. And for example, I had one task where I literally asked it to do something and it wrote one paragraph. And then, funnily enough, if you do struggle with GPT4, what I would say for you is to do this. And this has been my secret hack for actually getting great results out of ChatGPT. So I go over to ChatGPT and what I do is I click ChatGPT Classic. The reason I use ChatGPT Classic is because if we looked at the other GPTs, essentially the problem is that they have a huge system prompt and that essentially confuses the model. So if you want to get the very best out of GPT4, I would always recommend using ChatGPT Classic because the other one, like, if we go to just standard ChatGPT and GPT4, this has Dall E browsing and analysis, and it has so many different prompts and things inside of it that it does get very confused sometimes. So I would always recommend using ChatGPT Classic because it is far more effective at giving you results. So with that being said, are you guys excited for GPT 4.5? Are you excited that OpenAI is doing the gradual release? Are you excited Also for GPT 3.5? And are you going to be testing the new benchmarks to see exactly how it's happening? Are you also excited for the new future features, potentially, of having conversations with multiple GPTs? Enjoy the video. I'll see you in the next one.