I have to agree. Is it an F-U to US …maybe but I don’t care if running it locally. And honestly there is always competition as soon as the smartest newest model comes out. Feathers are just ruffled because they are outsiders .
We didn't get greedy. Consumers are greedy by giving out their data to China for free. We're putting our customers first and have greater privacy measures to justify the cost.
@@francisco444It’s greed when they created a walled garden stifling innovation and became rent seekers while demanding billions from tax payers. Deep Seek blew them out of the water.
No privacy concerns when giving our data to American big tech, let's give them everything, grant access to everything without even questioning and focus on how shinny and convenient their new tools are. But when it comes to the competitors...
@@ezequieligomez2135 there a fundamental good vs bad in this equation. Companies that at worst want to use your data to further their capitalistic goals, vs a one party authoritarian state who wants to use your data to generate more effective propaganda, influence elections and your opinion on important policy, all while whitewashing the massacre and unjust imprisonment of their own people. If you can't see that well then they've already got their hooks in you.
Sam Altman became a billionaire with OpenAI. It was meant to be open and free, instead he closed the code and added a paywall. I am glad about this competition with Deepseek, it's a lesson in humilty.
Fu... I just did it. I ran the smaller model in ollama on my machine and got pretty fast response although it was limitied in its replies. Then I tried the full fat model and whilst a bit slow in typing out the replies it was still very solid. I have a 16GB 5 year old laptop with admittedly an early i9 processor but only a GTX 1650 graphics card, which didn't seem to be used much. It ran with the CPU and it did not set the fans whirring very much at all!! That is legitimately bonkers! 🤯
To clarify, the 8B model - yeah - I did overhype the biggy. Apologies. Still, a seriously better response between the 1.5B and then 7B, and the 8B is a little better yet. I think it is time to buy more laptop memory :)
I love this channel for the fact that it's so down-to-earth and not seeking out viewers through clickbait and sensation. I have a disdain for those kind of video's that really try to grab your attention through dramatic titles and thumbnails always seeming to suggest in every of their video's that something shocking has just happened as to entice you to click on that video. And often the video itself is also unreasonably dramatic This channel isn't like that, and that's why I stick around here as pretty much the only AI-update content creator that I watch (not that there aren't other content creators who don't clickbait, but still. I love this channel for other reasons too). Just a reminder Igor to you and your team that there are people who really appreciate the sober, non-dramatic, down-to-earth content and stick around for that reason 😉. Hmmm... I could've probably used ChatGPT to articulate this comment better... Oh well 🙃
Awh beautifully said, thank you. But I do have to say that we are also guilty of clickbaiting with titles/thumbnails! Unfortunately it's just a part of RUclips that is impossible to avoid but I hear your point as we bring some amount of self regulation to the channel whereas some others would cross every line if if helped get 20 % more clicks. The videos itself are a different story though! My goal is to be your tech obsessed friend that catches you up on what is happening and how to use this stuff to do better. Again, thanks for the kind words. It's appreciated.
@aiadvantage Yeah I get it it is more or less almost an inevitable part of RUclips, but the clickbait never feels over the top for me personally. Many content creators in my viewpoint basically sell their soul and their sense of honour just to get many people clicking on their video's as they can (which I don't think works very well usually to build a loyal fanbase. But maybe I'm wrong)
What an incredible channel you host ... always so clear, concise, and substantive .... I'm an older person who lacks amazing digital skills, but I am able to see the vast scope of AI that is birthing before us, and you make it possible for someone like me to understand the transitions that are happening in warp speed time ... thank you for the great vids ... "TOP-SHELF" ...
For the Operator part, I totally agree that many people still don't know its potential and think of it as a flight-booking/restaurant-reservation agent. I am not even using Operator. I am using a free Browser Use, and it works scaringly good.
Thanks for the video. For the privacy part, do you really believe big tech US companies don't do the same? Whatever we upload/input into ChatGPT, Gemini, etc... it's also used for their own benefit, we shouldn't be naive
The difference is that US Companies uses your data to their advantage, while Chinese companies collect data for CCP' advantage. We have laws in the west limiting the use of stolen IPs/data. China WILL use your stolen data/IPs and definitly not in your interest.
@@tobiashaurum3611 bullshit! we all know that tech companies already sold data to china, especially google it’s their main source of profit. They couldn’t care less.
@@tobiashaurum3611 I believe that Western companies directly use my data to ruin my society. I don't believe the Chinese government particularly cares about me or my data.
Excited to see the video about your new cases for Operator! Some of what you mentioned is really intriguing, like having Operator research multiple websites, organize the data, and even create a presentation without user input. I’m curious to see how that presentation looks-it’s probably still pretty basic and in need of some refinement, but the potential is there. On the topic of ordering groceries with Operator in one click, I do something similar without it. With Amazon Fresh, I can set up a subscription for delivery every two weeks or reorder my last delivery using the app in under two minutes. So, I’m not entirely sold on the convenience examples you gave for ordering food or making reservations, but I get that this is just the beginning as you’re still experimenting with user cases. Looking forward to seeing more of how Operator evolves and how you refine these features. Keep the updates coming-I’m excited to see what’s next!
I will never not be overjoyed at Sam looking bad - He is the villain in the AI story and we will reflect on his mistakes and sociopathic ego for the rest of humanity
I’ve been using deepseek R-1 Lite since before November 20 and 2.5 MOE for over 4 months! I don’t know where people get this information that it just came out? I feel like I live on another planet or this is just another example of the Mandela effect? Anyone who was actually deep in AI space, not just on a trend, saw this coming!
r1 came out last week but yeah it’s weird as a model at this level has bren available for months with o1 and now everyone goes crazy. it really is similar to gpt3 vs chatgpt
The weird thing is that some people think this hurts compute, we won't get less compute, we'll just do more AI. Just like increased productivity of human workers due to tools didn't make us work one day per week, but instead do more output.
Brilliant channel. No BS or clickbaiting or fake end of the world panic. Subscribed. 👍👍👍👍 To the point and easy to understand for people who aren't really into all this AI craze.
Been playing with it, must say it's good! Answers in context of previously asked questions. Asked to design a Transmission DWDM 100g link, and IP end to end configs, it layed out everything + diagrams. Excellent 👌🏽
@sara_phoebe If the Military-Industrial Complex is paying for AI verification advocacy, sign me up! Seriously though, Do you get paid for nationalistic comments? edit: This conversation isn't productive. I'm out.
Hi Igor, thanks for you thoughts. I used to follow a lot of AI channels but I converged to your channel as one of the few channels If now follow. You provide non hyped, solid information. No clickbaiting, hyping, just your educated thoughts on the subject. Thanks for that! much appriciated
Genuinely, has anyone got any tips to get the people who don't care or are scared about AI, into it? "Tell your loved ones about... Deep seek, operator etc" hit me hard because I'm in the UK and nobody cares and/or hates everything to do with it. Talking about AI in my country, feels like talking about death, nobody wants to do it.
Uff tough one honestly. Maybe having 5 mac minis with M4 chips that are tethered or like 6 x 5090s in a rig with 512 gb ram? All just estimates there though I don't exactly know but it would probably cost north of 15 k $. Again just my quick guess based on what a 400B models takes.
Their terms and conditions are clear. Nobody is twisting Yankees arm to use deepseek. Yankees can continue to pay and use their own AI. Any criticisms is out of malice and jealousy.
I have a feeling that in practice the price difference will not be as significant, the verbosity of the model's "thinking" is counted as output. Many will learn the hard way that CoT models are a tool that should be used when needed and no more.
00:28 well, you just described the entirety of the internet...And as for the data, how can you know that chat gpt doesn't do the same thing? Also, are you aware of a company named Adobe? Can you at least admit that they embarrassed OpenAI for a few million dollars? Doesn't THAT tell you something about market prices? Paying 200 dollars for something that is worth 15? That's a free market for you, enjoy it. I am not quite sure what you are complaining about in this video. Although, your "the thing that really matters" section of your video paints a pretty clear picture for me about your intentions for this video.
Thanks to You Mate, I have realistic expectations for this bullrun. Got into crypto early this year and I have truly learned a lot from you about cycles, crypto and markets as a whole. Now I know that all assets/markets are heavily tied to each other and global economy has very high impact on Crypto. Thats why I am very cautious about when traditional markets top and how this affects Bitcoin and altcoins. If i havent discovered your channel I would have blindly holding instead of engaging the market........ I have managed to grow a nest egg of around 2.3Bitcoin to a decent 24Bitcoin. in the space of a few months... I'm especially grateful to Evelyn Mya’s, whose deep expertise and traditional trading acumen have been invaluable in this challenging, ever-evolving financial landscape.
The pump and dump has been a perfect eyeopener for us all to really see how unpredictable the market can be and the need for us to be trading not just seat and hodl.
Thanks for the video. I appreciate that you've been pointing out for a while now when a model is chinese. I would also appreciate learning more about censorship, bias, and, last but not least, the speculations surrounding security. I think this kind of education is needed...
Good video and I totally agree with you! Operator is the game changer and it's going to help people doing more. I can specially see a big benefit for those who are not "high achievers", the common guy/girl. I believe that with this tool, there are a possibility for everyone to become a high achiever. 💪🏻😊
Do we think that DeepSeek will add a voice interaction with R1 so that you can not only interact with voice prompts and listen to it go through its thinking process?
Isn’t it cheaper to run deepseek on a server than to use the api from some provider. Because if you pay for the server by hour or something then you can api call as many times as you want
How release of DeepSeek model is related to nVidia stock price? Every model what users run locally are potential need more nVidia products.. stock price should go up.
Now there's likely going to be far less demand for their swankiest newest most costly gpus, hence far less profit. Meaning current valuation is bloated.
@@MuantanamoMobile They already have order books full, and multiple other areas where they will bring a new producs in the future. Stock price is down its good opportunity to buy now some more nVidia stocks,
Thanks for this video. It was such a surprise to see the news on the stocks this morning. What exactly made it trigger today since the model had been out for a while. Loved hearing your perspective.
Here’s a thought. It always hallucinate. There’s nothing intelligent about the model that grounds it in reality other than through tokenized language. That isn’t enough for it to experience all of reality therefore it will hallucinate even right answers
I haven’t seen you this enthusiastic and overwhelmed in a long time. Your excitement in this video is truly contagious. It’s amazing to see! Greetings from Germany.
Just want to add that the rule of thumb to not attach or write anything you don’t want others to know about applies to all models, including ChatGPT et al.
What I'd like to know is whether China has a much more advanced AI that is not available to the public, but they have just matched OpenAI's capabilities for public purposes.
Excellent overview! I agree with most you're saying. But! Have you seen convincing documentation that DeepSeek's development costs are truly 5% of what we had otherwise been told this kind of thing costs (money; GPU)? I don't question DeepSeek as a sharp competitor (nor do I ignore it's open source), but is it really qualitatively a different beast on the development side of things?
Great point. I trust bog AI voice like Emad on this saying its realistic but I guess we will find out very soon where every lab on earth is going to try to replicate their published findings.
It's very entertaining seeing the millionaires shook so bad by this. We've been in a state of "good enough" for too long. Our expectations need to be higher. Thank you china
Can someone explain what's the "trick" with DeepSeek, why did it cost way less to build? I've heard it in another video that basically it was not trained with data directly from the internet, but rather through the already exitisting gpt models? Meaning, they saved tons of money and computing power by training it through apps/tools like chatgpt. If this not true, what did the big tech companies miss, and what did Deepseek nail? Thanks :)
Let me ask you this I know that we can download these models and run them locally, but there’s something about the top model being pretty big and requiring a lot of computational power from our computer is this true. It’s not like would we be downloading the smartest or more powerful model or is it a water down version which is still good but not as good as a oh1 pro or not one pro just o1
Is it just me or does @deepseek_ai R1 uses @OpenAI structure in the background and if so, to what extend? I was able to see an OpenAI reference in the thinking process. R1 responded with "server busy" responses with multiple times and i simply asked if it was caused by something i did, then he think that it may be caused by OpenAI. The sentence from thinking process: "Possible causes could be high traffic on OpenAI's servers" please, somebody enlighten me :D
Operator and agents like it will change everything. Any online business can be created and run by agents. You’ll just need to coach them through parts and strategy. Landing page creations, email flow , ad creation, sales outreach system. Business like ecom store can be created easily. And optimized very well so well wit the data and limitless number of A/B tests it can do
Dude, the model that beats Open AI is about 480GB and very few people will be running it locally, unless they are racked up with GPUs. The local install Deepseek models are poor you are better off with Meta’s Ollama locally. The cloud version that has your data yes is also the Open Ai equivalence but the yanks are already hacking access to its use.
@@aiadvantage Its funny they call it a "computer" use agent yet it cant use your computer, only a virtual browser. Imo UI-TARS shows much more promise.
@FTBAngel oh it may be up for the moment but good luck trying to get any serious work done when every five minutes is out or it slows to half a token a second. The Chinese cannot scale this. And on top of that trying to serve it to the world for free and to all the billions of ppl in third world countries who have never had access to a model like this for work starting to hammer it and without rate limits? Heck US companies who actually have the compute required can barely scale a model world wide and you expect china to do it, and not only that but reliably? It's like buying a toy out of a 25 cent machine and expecting it not to break in 10 seconds. The economics are not there. Even if their claimed training cost is legit (doubtful), the economics of test time compute and inference are inescapable.
since it's mainstream, it's almost always down or busy and cannot answer any more even when trying hard ... so it starts to become useless .. free but useless and unusable for now until the architecture is scale up to absorb the high demand due to hype
As a consumer, we all want a great product at a great price. Keep in mind, there’s no such thing as a free lunch. If the offer sounds too good to be true, it usually is.
I do still still think ChatGPT is better due to the features you get other than just the o1 model, like GPTs, tooling, Canvas, AVM, etc. But it's definitely a wake up call to OpenAI to stop being too greedy as it felt so especially when they made a pro plan, and reduced the thinking time for o1 on plus plan, sometimes leading to worse outputs because it tries to think too quick when it shouldn't (maybe acceptable on o1-mini but not full o1) as far as sometimes being worse than o1-preview...
imagine when openAI implements the 50 times more efficient inference compute of r1.into o1/o3, the log scale AI intelligence increase should increase raw IQ of openAI model at least a sigma or 2 beyond r1 will be interesting to see openAI's next big thing. In some sense they've been gifted a grand opportunity to improve their world leading models even further and race forward to AGI which is what matter much more than cheap non AGI models
Does anyone know why I cannot get the sign-up code to my gmail address? I keep hitting 'send code' but I cannot find their eMail anywhere, not even in my spam folder. HELP
![I.N "HALLUCINATION" | [Stray Kids : SKZ-PLAYER]](http://i.ytimg.com/vi/n5B5q1Hwt_U/mqdefault.jpg)
![[1hr Talk] Intro to Large Language Models](/img/1.gif)
Today the consumer wins
What about running it locally? How safe it is?
@@JDSpartan2007 100 % safe. If you install it locally no data leaves your computer. That's why this is so crazy.
I have to agree. Is it an F-U to US …maybe but I don’t care if running it locally. And honestly there is always competition as soon as the smartest newest model comes out. Feathers are just ruffled because they are outsiders .
It’s funny because Warren Buffett had just put out something about this being the year of the little companies…
@@aiadvantage That is crazy! Thank you!
Glad that OpenAI finally got humbled. They got too greedy.
free market doing it’s thing. love to see it
They deserved to be humbled the moment they closed openai imo.
how is it greed when they're not even profitable?
it takes a lot of money to support 300M weekly active users. 1B supposedly by end of 2025.
We didn't get greedy. Consumers are greedy by giving out their data to China for free. We're putting our customers first and have greater privacy measures to justify the cost.
@@francisco444It’s greed when they created a walled garden stifling innovation and became rent seekers while demanding billions from tax payers. Deep Seek blew them out of the water.
US : you need Ferrari to run AI
China: No, a potato 🥔 is enough
ahaha I love this comment
If a potato can become vodka, I can become anything
DeepSeek is a game changer. A playing field leveler. Local host. Open source. Beautiful. 🔥
How exactly are you going to host localy a model that requires 720 GB VRAM? (671B parameters)
You won't need 1 billion dollars like how American companies were saying xD
No privacy concerns when giving our data to American big tech, let's give them everything, grant access to everything without even questioning and focus on how shinny and convenient their new tools are. But when it comes to the competitors...
@@ezequieligomez2135 there a fundamental good vs bad in this equation. Companies that at worst want to use your data to further their capitalistic goals, vs a one party authoritarian state who wants to use your data to generate more effective propaganda, influence elections and your opinion on important policy, all while whitewashing the massacre and unjust imprisonment of their own people. If you can't see that well then they've already got their hooks in you.
Same story with TikTok. They will fear-monger around anything that threatens them from having control and monopoly.
I just left a comment under this comment but I can’t see it? RUclips’s censorship has been increasing.
@@Sxd212 come on! what is this? Chinese misinformation? 😂
Yeah, I noticed some of my comments on different topics on YT were removed.
it depends on how much of a toxicity/sensitive a word inside a phrase carries now . i heard something regarding yt new policy .
Sam Altman became a billionaire with OpenAI. It was meant to be open and free, instead he closed the code and added a paywall. I am glad about this competition with Deepseek, it's a lesson in humilty.
It's a big kick in the butt to every middleman, including Altman.
Fu... I just did it. I ran the smaller model in ollama on my machine and got pretty fast response although it was limitied in its replies. Then I tried the full fat model and whilst a bit slow in typing out the replies it was still very solid. I have a 16GB 5 year old laptop with admittedly an early i9 processor but only a GTX 1650 graphics card, which didn't seem to be used much. It ran with the CPU and it did not set the fans whirring very much at all!! That is legitimately bonkers! 🤯
This! Thanks for sharing
Wym full model. The 600b one?
@@MartinsBaklansJr. nono probably the 7b one
I believe the full model requires 1,280GB VRAM, so no way consumer hardware will do.
To clarify, the 8B model - yeah - I did overhype the biggy. Apologies. Still, a seriously better response between the 1.5B and then 7B, and the 8B is a little better yet. I think it is time to buy more laptop memory :)
I love this channel for the fact that it's so down-to-earth and not seeking out viewers through clickbait and sensation. I have a disdain for those kind of video's that really try to grab your attention through dramatic titles and thumbnails always seeming to suggest in every of their video's that something shocking has just happened as to entice you to click on that video. And often the video itself is also unreasonably dramatic
This channel isn't like that, and that's why I stick around here as pretty much the only AI-update content creator that I watch (not that there aren't other content creators who don't clickbait, but still. I love this channel for other reasons too).
Just a reminder Igor to you and your team that there are people who really appreciate the sober, non-dramatic, down-to-earth content and stick around for that reason 😉.
Hmmm... I could've probably used ChatGPT to articulate this comment better... Oh well 🙃
Awh beautifully said, thank you.
But I do have to say that we are also guilty of clickbaiting with titles/thumbnails! Unfortunately it's just a part of RUclips that is impossible to avoid but I hear your point as we bring some amount of self regulation to the channel whereas some others would cross every line if if helped get 20 % more clicks.
The videos itself are a different story though! My goal is to be your tech obsessed friend that catches you up on what is happening and how to use this stuff to do better.
Again, thanks for the kind words. It's appreciated.
@aiadvantage Yeah I get it it is more or less almost an inevitable part of RUclips, but the clickbait never feels over the top for me personally. Many content creators in my viewpoint basically sell their soul and their sense of honour just to get many people clicking on their video's as they can (which I don't think works very well usually to build a loyal fanbase. But maybe I'm wrong)
What an incredible channel you host ... always so clear, concise, and substantive .... I'm an older person who lacks amazing digital skills, but I am able to see the vast scope of AI that is birthing before us, and you make it possible for someone like me to understand the transitions that are happening in warp speed time ... thank you for the great vids ... "TOP-SHELF" ...
For the Operator part, I totally agree that many people still don't know its potential and think of it as a flight-booking/restaurant-reservation agent. I am not even using Operator. I am using a free Browser Use, and it works scaringly good.
Thanks for the video.
For the privacy part, do you really believe big tech US companies don't do the same? Whatever we upload/input into ChatGPT, Gemini, etc... it's also used for their own benefit, we shouldn't be naive
Yes, exactly!
The difference is that US Companies uses your data to their advantage, while Chinese companies collect data for CCP' advantage. We have laws in the west limiting the use of stolen IPs/data. China WILL use your stolen data/IPs and definitly not in your interest.
@@tobiashaurum3611 bullshit! we all know that tech companies already sold data to china, especially google it’s their main source of profit. They couldn’t care less.
@@tobiashaurum3611 I believe that Western companies directly use my data to ruin my society. I don't believe the Chinese government particularly cares about me or my data.
Excited to see the video about your new cases for Operator! Some of what you mentioned is really intriguing, like having Operator research multiple websites, organize the data, and even create a presentation without user input. I’m curious to see how that presentation looks-it’s probably still pretty basic and in need of some refinement, but the potential is there.
On the topic of ordering groceries with Operator in one click, I do something similar without it. With Amazon Fresh, I can set up a subscription for delivery every two weeks or reorder my last delivery using the app in under two minutes. So, I’m not entirely sold on the convenience examples you gave for ordering food or making reservations, but I get that this is just the beginning as you’re still experimenting with user cases.
Looking forward to seeing more of how Operator evolves and how you refine these features. Keep the updates coming-I’m excited to see what’s next!
Fyi...ai agent's (via computer agents) on abacus can do that too.
I will never not be overjoyed at Sam looking bad - He is the villain in the AI story and we will reflect on his mistakes and sociopathic ego for the rest of humanity
I really appreciate your comments different from the hype of the many many other RUclipsrs…
I’ve been using deepseek R-1 Lite since before November 20 and 2.5 MOE for over 4 months! I don’t know where people get this information that it just came out? I feel like I live on another planet or this is just another example of the Mandela effect? Anyone who was actually deep in AI space, not just on a trend, saw this coming!
r1 came out last week but yeah it’s weird as a model at this level has bren available for months with o1 and now everyone goes crazy. it really is similar to gpt3 vs chatgpt
It's the Davos effect - Nadya Satella mentioned it, as did a couple other folks.
The weird thing is that some people think this hurts compute, we won't get less compute, we'll just do more AI. Just like increased productivity of human workers due to tools didn't make us work one day per week, but instead do more output.
Brilliant channel. No BS or clickbaiting or fake end of the world panic. Subscribed. 👍👍👍👍
To the point and easy to understand for people who aren't really into all this AI craze.
thank you and to the team so much for this talk.
Been playing with it, must say it's good! Answers in context of previously asked questions. Asked to design a Transmission DWDM 100g link, and IP end to end configs, it layed out everything + diagrams. Excellent 👌🏽
Most relevant, and informative, conclusion I've heard yet. Thank you!
It was just added to Perplexity if you are worried about the use of your data
This is to mean? Data uploaded on perplexity is equally subject to the same privacy policy?
We need an independent party to verify their training methods and replicate their results on a budget.
Already happening as they shared it all in the paper that came with this.
@@aiadvantageHave they released the training data?
Huggingface is trying to replicate R1.
@sara_phoebe Reflection 70B! Big claims in AI always need independent verification. Do you get paid for nationalistic comments?
@sara_phoebe If the Military-Industrial Complex is paying for AI verification advocacy, sign me up! Seriously though, Do you get paid for nationalistic comments?
edit: This conversation isn't productive. I'm out.
Competition is healthy for companies. Allows us as users benefit as a result 👌🏻
Hi Igor, thanks for you thoughts. I used to follow a lot of AI channels but I converged to your channel as one of the few channels If now follow. You provide non hyped, solid information. No clickbaiting, hyping, just your educated thoughts on the subject. Thanks for that! much appriciated
Thank you for another great video. I had to come directly to your channel to get trustworthy and objective information on Deepsake.
Please continue creating videos about Operator, as I’m not yet comfortable assigning it sensitive tasks at this time.
this is business karma. OpenAI is being punished
Love that it's competition to keep everyone honest
Genuinely, has anyone got any tips to get the people who don't care or are scared about AI, into it? "Tell your loved ones about... Deep seek, operator etc" hit me hard because I'm in the UK and nobody cares and/or hates everything to do with it. Talking about AI in my country, feels like talking about death, nobody wants to do it.
I'd love a step by step vid on how to download and run Deepseek locally
“It ain’t about the money.” - Xi Jingping.
What stats are required to get the 671b model running locally?
Uff tough one honestly. Maybe having 5 mac minis with M4 chips that are tethered or like 6 x 5090s in a rig with 512 gb ram? All just estimates there though I don't exactly know but it would probably cost north of 15 k $. Again just my quick guess based on what a 400B models takes.
DeepSeek nibbled on All Might’s hair and delivered One For All for all of AI. 💪🏽
Their terms and conditions are clear. Nobody is twisting Yankees arm to use deepseek. Yankees can continue to pay and use their own AI. Any criticisms is out of malice and jealousy.
You weren't even looking at the correct model the whole time. It's r1 not v3 that we're looking at
Ah the github repo? You are right! Lol first person to notice thanks.
I made a similar comparison in my video. Basically, the consumer wins. Competition is good for us.
Best video I have seen on the topic so far, thanks!
Thank you , the vast majority of people just want the noise out and understand how it will affect everyday , everyday people
Great video. Thank you. Question: What's the difference between saving time and adding time to your day?
Love the one shot/no edit format.
I have a feeling that in practice the price difference will not be as significant, the verbosity of the model's "thinking" is counted as output. Many will learn the hard way that CoT models are a tool that should be used when needed and no more.
I listened to you and opened my mouth wide with your reservation example. Thank you 🙏🏼
00:28 well, you just described the entirety of the internet...And as for the data, how can you know that chat gpt doesn't do the same thing? Also, are you aware of a company named Adobe? Can you at least admit that they embarrassed OpenAI for a few million dollars? Doesn't THAT tell you something about market prices? Paying 200 dollars for something that is worth 15? That's a free market for you, enjoy it. I am not quite sure what you are complaining about in this video. Although, your "the thing that really matters" section of your video paints a pretty clear picture for me about your intentions for this video.
Fantastic video. Thank you!
Thanks to You Mate, I have realistic expectations for this bullrun. Got into crypto early this year and I have truly learned a lot from you about cycles, crypto and markets as a whole. Now I know that all assets/markets are heavily tied to each other and global economy has very high impact on Crypto. Thats why I am very cautious about when traditional markets top and how this affects Bitcoin and altcoins. If i havent discovered your channel I would have blindly holding instead of engaging the market........ I have managed to grow a nest egg of around 2.3Bitcoin to a decent 24Bitcoin. in the space of a few months... I'm especially grateful to Evelyn Mya’s, whose deep expertise and traditional trading acumen have been invaluable in this challenging, ever-evolving financial landscape.
I’ve seen many positive comments about Evelyn Mya. Can someone guide me on how to reach her?
SHE IS ON TELE GRAM.
@Evelynmya
The pump and dump has been a perfect eyeopener for us all to really see how unpredictable the market can be and the need for us to be trading not just seat and hodl.
Got a chance to benefit from her services a few weeks ago and it has been a very smooth experience.
The well timed drinks of water after a major point is golden 😅. Insightful content! Can you tell me whats the big deal with Operator though?
Deepseek was awesome the past couple of days, now the searches take minutes at a time. To be expected since it is becoming more popular i suppose
Thanks for the video. I appreciate that you've been pointing out for a while now when a model is chinese. I would also appreciate learning more about censorship, bias, and, last but not least, the speculations surrounding security. I think this kind of education is needed...
Thank you for another fantastic video!
Funny how tech people are still surprised by the China-US race.
And how political people are still surprised by AI's growth
You think operator or a competing tool could run an entire e-commerce business soon?
Good video and I totally agree with you!
Operator is the game changer and it's going to help people doing more. I can specially see a big benefit for those who are not "high achievers", the common guy/girl. I believe that with this tool, there are a possibility for everyone to become a high achiever. 💪🏻😊
Do we think that DeepSeek will add a voice interaction with R1 so that you can not only interact with voice prompts and listen to it go through its thinking process?
Isn’t it cheaper to run deepseek on a server than to use the api from some provider. Because if you pay for the server by hour or something then you can api call as many times as you want
How release of DeepSeek model is related to nVidia stock price? Every model what users run locally are potential need more nVidia products.. stock price should go up.
Now there's likely going to be far less demand for their swankiest newest most costly gpus, hence far less profit. Meaning current valuation is bloated.
@@MuantanamoMobile They already have order books full, and multiple other areas where they will bring a new producs in the future. Stock price is down its good opportunity to buy now some more nVidia stocks,
You have become my favorite RUclips channel! ✌️
I am running a ai RUclips automation company from my experience deep seek is the best ai model for youtube scripting it's on demon mode🔥
Thanks for this video. It was such a surprise to see the news on the stocks this morning. What exactly made it trigger today since the model had been out for a while. Loved hearing your perspective.
Also, DeepSeek makes a lot of hallucinations when I ask you about my favorite rhythm game players . AI is definitely still making up a lot of things
Here’s a thought. It always hallucinate. There’s nothing intelligent about the model that grounds it in reality other than through tokenized language. That isn’t enough for it to experience all of reality therefore it will hallucinate even right answers
So did Deep Seek just open pandora's box by making ease of use and utilization of AI available to everyone and their mom ?
Only downside so far I've encountered is the context window. Besides that, the responsed are way better than Claude and ChatGPT
I haven’t seen you this enthusiastic and overwhelmed in a long time.
Your excitement in this video is truly contagious. It’s amazing to see!
Greetings from Germany.
Very good explanation on Deepseek 👍
For a European these times, Is it worse giving your data to China than to an American oligarch?
Just want to add that the rule of thumb to not attach or write anything you don’t want others to know about applies to all models, including ChatGPT et al.
What I'd like to know is whether China has a much more advanced AI that is not available to the public, but they have just matched OpenAI's capabilities for public purposes.
I really like the way you presented this.
Thank you :) I think of it as sitting down over a coffee with a friend who asks "Ok so what's going on with this DeepSeek thing?"
Excellent overview! I agree with most you're saying. But! Have you seen convincing documentation that DeepSeek's development costs are truly 5% of what we had otherwise been told this kind of thing costs (money; GPU)? I don't question DeepSeek as a sharp competitor (nor do I ignore it's open source), but is it really qualitatively a different beast on the development side of things?
Great point. I trust bog AI voice like Emad on this saying its realistic but I guess we will find out very soon where every lab on earth is going to try to replicate their published findings.
China has an excellent economic model
Great POV. Awesome channel.
Great overview and perspectives on this. Thx man.
You can run DeepSeek locally without internet connection?
How compares DeepSeek R1 to Gemini Experimental 1206?
It's very entertaining seeing the millionaires shook so bad by this. We've been in a state of "good enough" for too long. Our expectations need to be higher. Thank you china
Can someone explain what's the "trick" with DeepSeek, why did it cost way less to build? I've heard it in another video that basically it was not trained with data directly from the internet, but rather through the already exitisting gpt models? Meaning, they saved tons of money and computing power by training it through apps/tools like chatgpt. If this not true, what did the big tech companies miss, and what did Deepseek nail? Thanks :)
Let me ask you this I know that we can download these models and run them locally, but there’s something about the top model being pretty big and requiring a lot of computational power from our computer is this true. It’s not like would we be downloading the smartest or more powerful model or is it a water down version which is still good but not as good as a oh1 pro or not one pro just o1
Is it just me or does @deepseek_ai R1 uses @OpenAI structure in the background and if so, to what extend? I was able to see an OpenAI reference in the thinking process. R1 responded with "server busy" responses with multiple times and i simply asked if it was caused by something i did, then he think that it may be caused by OpenAI. The sentence from thinking process: "Possible causes could be high traffic on OpenAI's servers" please, somebody enlighten me :D
i would be interested on you testing UI Tars vs Operator
Operator and agents like it will change everything. Any online business can be created and run by agents. You’ll just need to coach them through parts and strategy. Landing page creations, email flow , ad creation, sales outreach system. Business like ecom store can be created easily. And optimized very well so well wit the data and limitless number of A/B tests it can do
could they have buried in the open source code that does send the information whenever the local version connects to the web.?
I genuinely prefer to use Deepseek given that it is proven it's best performed
very helpful, thanks
very well done and great info. thx subbed
How is a closed-source AI company called? Yep, OpenAI.
Dude, the model that beats Open AI is about 480GB and very few people will be running it locally, unless they are racked up with GPUs. The local install Deepseek models are poor you are better off with Meta’s Ollama locally. The cloud version that has your data yes is also the Open Ai equivalence but the yanks are already hacking access to its use.
Please stop fearmongering about China. I say this as an American
Operator is literally what rabbit r1 promised and scammed millions 😂😂
Yeah totally. The idea is a powerful one but it turns out the execution is hard
@@aiadvantage Its funny they call it a "computer" use agent yet it cant use your computer, only a virtual browser. Imo UI-TARS shows much more promise.
You have access? Because for me it's been down all day 😂 whoops i guess scaling is hard, who knew.
Its up now, atleast
@FTBAngel oh it may be up for the moment but good luck trying to get any serious work done when every five minutes is out or it slows to half a token a second. The Chinese cannot scale this. And on top of that trying to serve it to the world for free and to all the billions of ppl in third world countries who have never had access to a model like this for work starting to hammer it and without rate limits? Heck US companies who actually have the compute required can barely scale a model world wide and you expect china to do it, and not only that but reliably? It's like buying a toy out of a 25 cent machine and expecting it not to break in 10 seconds. The economics are not there. Even if their claimed training cost is legit (doubtful), the economics of test time compute and inference are inescapable.
since it's mainstream, it's almost always down or busy and cannot answer any more even when trying hard ... so it starts to become useless .. free but useless and unusable for now until the architecture is scale up to absorb the high demand due to hype
Very exciting topic!
Deepseek took Metas & Alibaba's open source and added reasoning. We are giving them too much credit.
As a consumer, we all want a great product at a great price. Keep in mind, there’s no such thing as a free lunch. If the offer sounds too good to be true, it usually is.
LOCAL AI HAS TO WIN ! , if we don't own it, is not ours
If the news tell you it's bad, then it is good
Why everyone is scared of china. If you use internet it is world wide. ANything you input is on the internet.
Good times. Now we have space.
I do still still think ChatGPT is better due to the features you get other than just the o1 model, like GPTs, tooling, Canvas, AVM, etc. But it's definitely a wake up call to OpenAI to stop being too greedy as it felt so especially when they made a pro plan, and reduced the thinking time for o1 on plus plan, sometimes leading to worse outputs because it tries to think too quick when it shouldn't (maybe acceptable on o1-mini but not full o1) as far as sometimes being worse than o1-preview...
imagine when openAI implements the 50 times more efficient inference compute of r1.into o1/o3, the log scale AI intelligence increase should increase raw IQ of openAI model at least a sigma or 2 beyond r1
will be interesting to see openAI's next big thing. In some sense they've been gifted a grand opportunity to improve their world leading models even further and race forward to AGI which is what matter much more than cheap non AGI models
Does anyone know why I cannot get the sign-up code to my gmail address? I keep hitting 'send code' but I cannot find their eMail anywhere, not even in my spam folder. HELP
they got hacked that's why you are having issues to sign up and start using Deepseek. Need to wait a few days
They got hacked.
What about browser-use?
Great, great report.