# The OpenAI Team Just Revealed Whats Coming Next...

## Метаданные

- **Канал:** TheAIGRID
- **YouTube:** https://www.youtube.com/watch?v=2VSsfINqZf0
- **Дата:** 19.10.2024
- **Длительность:** 11:28
- **Просмотры:** 45,770
- **Источник:** https://ekstraktznaniy.ru/video/13967

## Описание

Prepare for AGI with me - https://www.skool.com/postagiprepardness 
🐤 Follow Me on Twitter https://twitter.com/TheAiGrid
🌐 Checkout My website - https://theaigrid.com/

00:00:00 - Introduction
00:00:28 - Model series
00:00:53 - New era
00:01:17 - Model classes
00:01:42 - Use cases
00:02:09 - Model integration
00:02:36 - Future intelligence
00:03:27 - Upcoming updates
00:03:50 - Model capabilities
00:04:19 - GPT-4 training
00:04:43 - Exploration capabilities
00:05:37 - Video summarization
00:06:01 - 3D synthesis
00:06:44 - Business implications
00:07:05 - Audio modality
00:07:49 - Image scenarios
00:09:30 - Search GPT
00:09:53 - Dynamic routing
00:10:37 - Model improvements
00:11:21 - Closing thoughts

Links From Todays Video:


Welcome to my channel where i bring you the latest breakthroughs in AI. From deep learning to robotics, i cover it all. My videos offer valuable insights and perspectives that will expand your knowledge and understanding of this rapidly evolving field. Be sure t

## Транскрипт

### Introduction []

so openai have released a very fascinating webinar in which they talk about the future of 01 models they show us a few graphs and gives us quite the insight into what is coming in the future now this isn't the first time that open has done a webinar in which they showcase the potential for future models but I think this one is rather fascinating as it allows us to look at where the new series of models is headed

### Model series [0:28]

as you may have recently realized open and I have released the recent model o01 and in doing so they've created a separate model series in relation to the GPT Series so you can see right here on this graph this is what I'm talking about this is where we can see that not only do we have the GPT 4 Series in addition we now have the 01 series as well as the GPT series you can see right

### New era [0:53]

here that we have these two model series that are going to be continually released and this basically ushers us into a new era of AI because this now means that unlike prior Cycles where we would get just GPT 3. 5 GPT 4 and although right now there are rumors of potentially GPT 4. 5 this is just going to be in the GPT Series so it seems that

### Model classes [1:17]

we're going to be getting the 01 series and then of course the GPT series as two separate model classes one model class which can think and reason for very long periods of time whereas we're getting the other series of models which is going to be a more round General use model for everyday purpose and I think that makes a lot of sense considering the fact that these 01 series have tremendous use cases now in this private

### Use cases [1:42]

webinar they did talk about a bunch of use cases but I'll be covering that in another video because model capabilities are something that are really fascinating and the insights with as to what is going to be in the future is something I always like to look at open AI our product and research teams are deeply integrated and this close partnership means that with every model release we see new features and capabilities coming online in chat GPT

### Model integration [2:09]

you'll likely are you're likely familiar already with our GPT series the latest model in that series being GPT 40 and just last month we released a new type of model series called open ai1 we'll continue to release models in both model series because they help solve very different problems in use cases and in practice we see many customers use both models and that's how we use them internally for our own processes and use cases as well so I

### Future intelligence [2:36]

think this is something that's rather fascinating because if we take a look at this old graph you can remember that we were looking at model intelligence going up as well as the models called GPT next and future models in future years so I think this is something that we need to pay attention to because we now know that with model intelligence we're going to be getting two eras where we have the 01 series on this new graph and then of course the GPT series on this graph as well so both models over the coming years will be increasingly more intelligent and as she said we're going to be getting to see more capabilities come online and I think that statement was of course referring to the agentic capabilities of future models now there is also another part where they did speak about what exactly is going to be next in these models and I think one of

### Upcoming updates [3:27]

the things that they spoke about was really fascinating let's touch on what's next we are planning to make several updates to these models over the coming months we plan to continue developing and releasing models in the new open A1 series as well as our GPT Series in addition to model updates we expect to add web browsing file and image uploading and other features to make them more useful uh in use cases in

### Model capabilities [3:50]

chat GPT see there he said he plans to add well not he but of course openingi plan to add multiple different features to these models in order to make them a lot more interesting and give capabilities now one of the things that I think many people do consistently forget and that includes myself is that the GPT 40 model is largely one of the most capable models that currently exists but is only currently in certain modalities if you remember how GPT 40

### GPT-4 training [4:19]

was initially trained was that it was trained as an omn model meaning that you could in theory put any kind of input in and receive out now this is something that has slipped the minds of everyone because gp40 didn't ship with these capabilities initially which meant that people didn't realize that you could use it for other things

### Exploration capabilities [4:43]

now I'm guessing that the other features that are probably going to be added are the ones on this web page right here that I'm going to show you guys in a moment so now if you come over to this web page which is on the Hello GPT 40 web page and you scroll down all the way until it says exploration of capabilities you're going to see that there are a variety of areas where you can see the new capabilities that these models are going to be getting in the future now of course there's no promises but this was something that we saw gp40 can do remember this model was trained on a variety of different inputs and outputs that allows it to do anything one of the exploration capabilities that I think is going to be most useful is of course the video summarization you can see the user inputs a video of a presentation on techniques for maximizing llm performance and says can you give me a detailed summary of the

### Video summarization [5:37]

presentation interestingly this is a 45-minute video and you can see that the output manages to literally summarize this entire video I can't imagine how useful this would be for the average person if this capability is there now I don't think that the average person is going to use this necessarily on a day-to-day basis but adding certain capabilities is going to open up a wide range of different use cases Now video

### 3D synthesis [6:01]

isn't the only thing that gp40 might be getting in the near to short-term future if we look at many of the other things we can see here we've got also 3D object synthesis so we can say our input is a realistic looking 3D rendering of the opening iio logo shown below and you can see that the output is there then also there and then of course we get a 3D Reconstruction from six generated images so it's pretty crazy that we do have a model that can generate images and then use those images to generate a complete 3D rendering in a video format so I think these are the kinds of things that most people don't realize that the models can do and are going to do in the future now why is this important well a

### Business implications [6:44]

lot of people are banking on these models for future businesses future ideas and I think it's always important to understand the kinds of things that these models are going to be able to do in the future so that if you're trying to plan for your business for Content creation or just general daily use you can understand where these models are going to excel another example SL

### Audio modality [7:05]

modality that we do have there is of course audio for those of you that work with audio frequently it's quite likely that in the future Google's Gemini isn't going to be the only model that allows you to input audio as an input currently this model exists because it allows for hours and hours of footage but it seems that in the future open AI GPT 40 omn model is going to be the one that has these Fe features you can see that there is a minute audio input in and then someone asks how many speakers are in this audio and what happened and the output is that there are four speakers in the audio and it describes it and someone also asks GPT 40 to transcribe this and you can see that the model manages to easily transcribe exactly what's going on I think this kind of

### Image scenarios [7:49]

thing is going to be really useful in the future considering the fact that many times you have many different file types file sizes and it just allows you to immediately work with GPT 4/ the Omni model in a way that is much more seamless another one that I don't think most people will be using this but I still think it's an exploration of capabilities is where you can get one single image and you can use that image as a story in multiple different character scenarios you can see here we've got Sally the male woman and you can see that we can have this character in multiple different scenarios simply based on one picture and you can see that all you're doing is inputting these simple prompts and you're saying uh-oh Sally tripped and the dog is chasing her and you can see we immediately get an image that looks like Sally's tripped then you can see we just update the story and it turns out it was a nice dog and then we can see Sally is driving away this is something that you know a lot of models do struggle with character consistency is something that many are looking for a solution in terms of which application they can use and if opening i1 can truly solve this problem it's going to be a really good thing for a lot of individuals now when we actually look back at this slide right here we can see that one of the things they also talk about is access to tools and one of the things he did also mention was of course web search now that reminds me of opening eyes recent tool which unfortunately I still don't have access yet but that is of course the very Infamous search GPT which is basically quite similar to perplexities online tool where using the power of large language models you can use basically an advanced version of Google Search that allows you to search the internet in a way that's much more effective I

### Search GPT [9:30]

wouldn't be surprised if search GPT is natively integrated into chat GPT over the coming months along with some small agentic tool use that makes it a lot more usable and autonomous and while even today you able to switch between models in the same conversation like you saw in the demo we're working to enable chat tvt to automatically choose the right model for your given prompt so

### Dynamic routing [9:53]

that is where it looks like we're going to be getting a dynamic model routing which is basically where you have one entire system and you don't really switch between models but you simply input your prompt as you would and then the system chooses which model to route your query to so for example let's say I asked chat GPT what is 3+ 3 it's probably going to Route it to GPT 40 mini the smallest lightweight easiest model for simple queries but let's say I say how do I fix this 10-step problem in my business it's going to go to open A1 and Route the question there because it knows that it's going to need multi step reasoning and I think this is going to be something that allows once again for you to have a much more seamless experience in terms of model updates

### Model improvements [10:37]

we've already started to test and run evals and open 01 which is the next iteration in this re following This research preview compared to o01 preview users can expect more powerful reasoning model that is even better at coding and math as you can see in these EV valves where o1 represents the orange bar we look forward to making 01 available to customers as we build as it will build on the use cases we saw today and likely unlock more use cases with its powerful reasoning capabilities so it looks like when openai manages to finalize their compute and they manag to actually ship out the final version of o1 We're going to be getting a truly intelligent and smarter model that's going to be even more capable than the 01 preview that we have now so with that being said let me know if you guys are excited for this you

### Closing thoughts [11:21]

if you guys are excited for this GPT 5 although they didn't mention that in today's video but I do think openi has a lot to offer