at this point it's impossible to ignore deep seek the app has shot up to number one on the App Store charts it's caused panic in the stock market and now some people prefer it over chat GPT but let's not get ahead of ourselves just yet because in this video we're going to explain exactly what deep seek R1 is why it's causing people to panic and how to use this insanely powerful AI model on your iPhone and your Mac and trust me it is far superior to Apple intelligence it's not even really the same thing but I had to throw that in there okay so first off what is deep seek and why is everybody going crazy over it so first off deep seek is an AI company out of China that recently launched a new open-source AI reasoning model called R1 and a reasoning model is an AI That's designed for multi-step thinking and problem solving rather than just predicting text like a glorified autocomplete now what makes R1 so special is not only that it matches the performance of open AI cutting edge1 model which is their reasoning model which by the way that's locked behind a $20 a month subscription but deep seek does this for free there's no subscription this is completely free and open source once again we need to reiterate that open source is a massive deal because yes you can use the web interface you can go on there and type in you know and do everything with a 50 message a day limit but the big selling point here for for deep seek the big Pro here is that it's open source which means that they can run this locally on your computer without sending your data off to the cloud and especially what a lot of people are worried about sending their data off to China but another key Point here and the reason that the whole stock market sold off and the reason people were panicking is because it came out that allegedly this model only cost under $6 million to build and train however what most people failed to mention is that that number is just for one training so the total cost is likely much higher but still the total tally is going to be significantly less than what you know open AI for example has spent training their models but the real game Cher is when it comes to cost for consumers and for Enterprise because the API usage is significantly cheaper so R1 costs $219 per million output tokens while open ai's 01 model costs $60 per million output tokens that is a fraction of the price so again that is a massive win for the consumer and especially for Enterprise who are going to be building apps you know with this okay so before we hop on the computer and the iPhone and show you deep seek in action we need to talk about what makes deep seek so special what makes it unique and different from say a chat GPT 01 because we know that deep seek matches the performance of 01 across math coding and reasoning tasks but what makes it unique aside from the cost well the main thing is that it uses a combination of supervised fine tuning and reinforcement learning or RL to handle complex reasoning tasks meanwhile most other AI models rely a lot more heavily on just the supervised fine-tuning aspect with a lot less emphasis on RL or reinforcement learning for self-improvement okay so let's talk about how to use deep seek and kind of compare it to chat GPT Siri with apple intelligence and so on so we're going to start on the iPhone and then we're going to go over to our computer to show some more advanced use cases for deep seek so first off you can see if you search for deep seek or if you go to the charts it's going to be number one in every chart you can see it right here so it's currently on version 1. 0. 6 that will vary based on when you download the application but if you go in you will see this right here you will need to make an account I would recommend using your Apple account or just signing in with a Google account that's going to be the quickest way to get into here now keep in mind iOS and Mac OS apps run in a sandboxed environment meaning that they cannot access other apps data or system files without explicit permission so if you're worried about this app stealing your contacts or stealing your messages or anything like that you know that stuff is not going to happen in these applications because once again it runs in a sandboxed environment you have to explicitly give permission before any type of application can collect that data and as you can see it was not asked for this application now I'm not saying that you know you should trust China with all your data that's not not what I'm saying I'm just saying that downloading the app and using the app is not going to have any type of security impact or security risk to your data that's on your phone that's outside of the deep seek application so anyways once you're in here on your new chat you want to make sure to enable deep think R1 so that's off by default you want to make sure that you are using the R1 model so select that and you can also select web search which is something that even the 01 model cannot do okay so we're going to enter a prompt that takes some type of thinking and reasoning so it's about two trains leaving at certain times going at certain speed at what time do they meet and also I said explain how you arrived at the answer with each step so if we go ahead and tap on send we can see how Okay so we're going to turn off search so it didn't like that so let's go ahead and do this without search and once again the servers are going crazy so you might have some issues so anyways once you type that in you will see right here that it actually shows what it's thinking it shows the reasoning before the answer so this is called chain of thoughts or coot reasoning and this is really cool this is also something that makes deep seek unique is that it shows you exactly what it's thinking so it says okay let's see so there's a problem about two trains so on and so on and it says hm so one train starts an hour earlier than the other that means by the time the second train starts and so on so it uses really humanlike reasoning before it answers the question it actually gives you a peak behind the curtain as to what you know the R1 model is thinking before it gives you an output and once again it uses RL or reinforcement learning which means that if it makes a mistake it will catch the mistake and learn from itself you know before it gives you the answer and even if you tell it it's wrong it will also you know show itself thinking about what went wrong and how it can fix it which is really awesome so here we go it gives us the answer at 12:26 p.
m. now if we do the same thing with chat GPT on the iPhone if we go ahead into our type to Siri right here and we type in the same thing it will show that okay so it's starting a workout so you can see that this is clearly a lot better than Apple intelligence which it's not even the same thing like I said earlier but I still wanted to show that that was pretty funny so let's go into chat GPT the application because that's where we'll be able to actually use this we're going to go up here and go to the 01 model and we'll type in the same thing we'll enter the same prompt and you can see the comparison between the two so you can see that chat GPT thought for 8 seconds and then it gives you the answer right here so chat GPT was a bit quicker but it doesn't really show you the reasoning it doesn't show you you know kind of how it came up with that answer so it just says determining the meeting point and that's it it doesn't really show you as much as deep seek did okay so now we're moving over to our computer our Mac right here so you can see the interface on desktop so you can get to this by going to chat. deep seek.
com this is the website right here so it works pretty much the same as the mobile application it's just a little bit different UI and it's much easier to work with on desktop so you want to make sure sure to select deep think R1 and once again we'll type in a new prompt here so we're going to copy this into deep seek R1 and we're also going to do the same here with 01 with open AI so we're going to go ahead and enter right there and then we'll do the same on deep seek and once again you'll see the reasoning here the thinking behind deep seeks answer which I just love this it's really cool to read you could also copy this as well and you can kind of paste this into chat GPT if you would like so surprisingly deep seek finished this before Chachi PT did with the 01 model model so that was pretty interesting and another thing I wanted to mention is if I type in something like wrong and saying that it's wrong and it says it'll actually kind of show you how it tries to resolve a potential hallucination which a Hallucination is just when a AI you know model gets something wrong so it tells you you know it's it basically runs through how R1 is thinking about if it hallucinated or not and how to potentially get to the right answer where if I did that with the 01 model you can see it will think about it but it won't show you how it's thinking and as far as actually writing the code I asked to write me working code for a simple 3D ball floating in space with a smaller ball floating inside of it so chat GPT gave me the code for this 3D ball and I put it into an HTML online viewer and you can see that is the result from the 01 model so now let's see what deep seat gives me and how it Compares now what's cool is that you can run the HTML code straight from Deep seek so you cannot do that here in chat GPT so you can see there's nothing to run but when I go to deep seek we have a run HTML button right here and we can click on that and you can see that I do not need that online viewer and another browser over here I can just run it straight from the deep seek interface however you can see that the result is a bit better with chat GPT it makes it a bit more modern and it shows a ball inside of a ball whereas deep seek really just shows me one red ball kind of floating right there so I like the answer better from 01 but I do like the interface here from Deep seek better where I can run it straight from the platform straight from the web interface and the last test that I want to show as an example in this video is one that I'm going to be personally using on a daily basis and it's a writing task just simply using these models to help me write and come up with ideas so I said write a funny script on the upcoming iPhone 17 and deep seek actually gave me a far better script so first off it gave me the reasoning right here and if you go down you can see it shows the title and the setting and even says Tim apple as like a joke about you know Tim Cook and just through this everything is is more personable it's more humanlike it's more humorous than the result that we got with chat gpt's 01 so honestly if it was not for the advanced voice mode on chat GPT on the mobile app and if it were not for the custom gpts then I would be switching over to deep seek pretty much fulltime and running it locally on my machine but I'm going to be holding on to chat GPT just because of those two features I mentioned since they are not available with deep seek now I also wanted to to show you how you can run deep seek locally on your Mac or your PC you could do this for llama for any of these other ones as well so I'm personally using LM Studio on my Mac you can go there at LM studio.
