00:00Namaskar, a Chinese AI, Deep Seek, has kept the world.
00:06American tech companies and the U.S. stock market
00:09had such a bad idea that they hadn't thought about their dreams.
00:12The release of Deep Seek AI from the Chinese companies should be a wake-up call.
00:16The fact is that on January 20th, a Chinese research lab
00:20has launched its AI chatbot.
00:22His name was Deep Seek R1 and his research paper was published
00:27in which he told us that this chatbot is the world's most advanced
00:32chatbot, Maths and Reasoning, as well as benchmarks,
00:35the Open AI, the chat GPT-O1 model, Meta, the Lama,
00:40Google, the Gemini advanced, all in one way.
00:45Deep Seek performance is the best, the efficient and efficient
00:48and the best performance is the best.
00:52But the most important thing is that this is the use of our
00:54you and all of us is completely free of cost.
00:58The other way, Open AI, the chat GPT-O1 Pro model
01:02is used for $200 per month.
01:05It's not enough, the train cost is only 5.6 million dollars
01:09reports.
01:10The other American companies, open AI, Meta or Google,
01:15are all billions of dollars
01:16they are spending their own artificial intelligence models
01:19to create their own artificial intelligence models.
01:20After this, look at the end of the conversation,
01:25when the Open AI founder Sam Altman asked
01:28if the chat GPT-O1 Pro model can create
01:32the Indians in India?
01:34But if you want to build foundational models, how should we think about that?
01:37With, you know, not 100 million, but let's say 10 million,
01:40could actually build something truly substantial.
01:42So, they had very arrogantly answered that no one can do this without us.
01:46You can try it, but you're hopeless.
01:48It's totally hopeless to compete with us on training and foundation models.
01:50You shouldn't try and it's your job to try anyway.
01:54Today, Sam will be feeling hopeless for yourself,
01:56because after DeepSeek's launch of DeepSeek,
01:58DeepSeek's launch of DeepSeek, which is a Chinese-built chatbot,
02:19immediately rattled investors and wiped a staggering $1 trillion of the US tech index.
02:24DeepSeek's launch of DeepSeek, which is a Chinese-built chatbot,
02:28now with exhaust profit,
02:31but now in a few days,
02:34the million-dollar won.
02:35This company is computer chips.
02:37These companies are specialized objects to own your computer chips.
02:49This is immediately after the Called activist price market.
02:54The $11 billion shares are 1 most expensive.
02:58Nasdaq is 3.1% below, but especially NVDA, why is it so bad?
03:04This is a very interesting cause, which we do now.
03:07But before we know about Deep Seek's story behind Deep Seek.
03:11Where did AI come from, and why did it surprise the world?
03:19Deep Seek's credit for creating a credit for a 40-year-old Chinese entrepreneur, Liang Wenfeng.
03:25Look at the face of the face, because the face of the face is not visible.
03:29Till date, they have very few public appearances.
03:32They have their identity and their identity.
03:35Their life and history also has the information publicly available.
03:39But we know that in 2015, they have a high-flying hedge fund.
03:44Mathematics and AI used to invest in investments.
03:48In 2019, they have a high-flying AI foundation for artificial intelligence.
03:53But in May 2023, they have a power to build their own hedge funds,
03:59a side project to create a design.
04:03This is an AI model that they wanted to create a design.
04:04Liang has said that they could create an AI model.
04:07They wanted to create a better design than the world.
04:10And behind it, the scientific curiosity had been created.
04:13The profit was made and the payee needed to create a design.
04:15To create this AI model, when they started their research team, they didn't hire engineers.
04:22They hired a PhD student of China's top universities.
04:26They used their AI model to train the world's most difficult questions.
04:31Just within 2 years, they launched DeepSeek R1 model.
04:45The answer is, less than people thought. You don't need as much cash as we once thought.
04:50Only 200 people involved were involved in the building, and 95% of people were less than 30 years.
04:55Compare this to OpenAI company, which had more employees hired by 3,500 employees.
05:00Today, DeepSeek is a famous AI firm in which, Baidu, Alibaba, and ByteDance were not paid for tech giants.
05:08Now, the architecture of its architecture is a chain of thought model.
05:14Weak Open AI is a chain of thought model.
05:18The naming of the name is very confusing,
05:21because ChatGPT has a lot of strange names in their models.
05:24They also have a lot of unique names.
05:26Let me tell you a little bit about understanding.
05:28ChatGPT first, the public launch was GPT-3.
05:31In November 2022, then March 2023, they launched GPT-4,
05:36which was much better than GPT-3.
05:39Then, May 2022, GPT-4-O, which was a multi-model version.
05:45You could not only talk about text, but you could talk about it.
05:49You could also send photos.
05:51You could also send photos.
05:52You could also understand photos.
05:54You could also generate photos.
05:56Here is the name of multi-model.
05:58Then, September 12, when we launch our open AI a new model was on the O1 model.
06:03The first model of the chain of thought was on the main model on the chain of thought.
06:05The first material is the main model.
06:10It means that when you have a question generated at the beginning, you will be able to answer it.
06:15It was counter-question to me, whether it was the answer to the user, or whether it was better.
06:21One answer to some angles, then you will be able to answer it.
06:26This is the chain of thought.
06:28The reasoning process of human reasoning is to copy.
06:31For example, if you ask in the chat GPT-40, which one is bigger?
06:37It is the answer to the question.
06:42But the answer is the answer to the question.
06:46When you ask in the chat GPT-01, you will be able to answer it.
06:53My case is 18 seconds ago, the answer is the answer.
06:58The answer is the answer.
07:00This answer is the answer.
07:03The answer is the answer.
07:08This is the reason why Oven and Deep Seek models are advanced reasoning models.
07:15Logically, this is a good thing.
07:18Deep Seek is also a good thing in detail.
07:22Step by step, exactly what you think about it.
07:25Give me one truly original insight about humans.
07:29Let me think of something that makes humans unique.
07:33But those are pretty well known.
07:37The user needs to be something original.
07:39The traits can interact with the other way.
07:43Paragraph by paragraph, if you look at the answer,
07:46you can see the video on the point.
07:49You can see the answer from this question.
07:53You can see how many different possibilities are.
07:59This is the actual answer.
08:02You can see the answer from this question.
08:04This is the answer from this question.
08:06You can see the answer from this question.
08:08This is the answer from this question.
08:10This is the answer from this question.
08:12This is the answer from this question.
08:14This question is the answer from this question.
08:16You can see the answer from this question.
08:18The answer from this question.
08:20If you will find a number,
08:22you will answer it.
08:24of thought wale models se agar aap yeh karne ko bholo ge
08:27to ismei bhi woh kaafi samei tak
08:28suchte rehte hain. Yeh screen recording
08:30dekhiye bichara deep seek isi baat
08:32meh ulaj gaya hai ki kaunsa number
08:34diya jaya. Kya mein koji commonly istimal
08:36kye jane wala number dhe dhuun yaa phir
08:38zero ya one ke bich mein koji number dhuun.
08:40Mein 42 kehsakta houn lekin
08:42woh kehna shayad cliche hooga yaa phir
08:44mein 17 yaa 73 choose kar sakta houn.
08:46Kya mein 1 se 100 ke bich mein
08:48hi number select karou yaa us range ke bahar
08:50jauun. Kyaunki user ne to completely random
08:52kaha hai. Agar mein ho karun ga to
08:54yeh completely random nahi hooga.
08:55Bilkul ek overthink karne wale insaan ki taro
08:58hai. Lekin finally jab yeh jawaab
08:59deta hai. To yeh yeh bhi kehta hai ki
09:01meinne apni taraf se socha ki 1 se 100
09:03ki range rakh li jayai. Or phir
09:05meinne 42 ka number chuna.
09:07Viesekar aap isi tariqe se or in-depth
09:09knowledge lena chate hai in AI chatbots
09:12ke baare mein. Or apne aap ko is field me
09:13upskill karna chate hai. To meinne pura
09:158.5 ghante ka AI chatbots ko master
09:17karne ka course bana ya hai. Pahle
09:192 chapters theoretical knowledge pere focus
09:21karte hai baaki ke 6 chapters practical
09:23advice pere. Students, employees, business
09:25owners ghar mein iska kiis se istimal kiya
09:27jasaakta hai. Latest lessons mein
09:29aap ko isme chat GPT visions, image
09:31generation daily 3 ke saath advanced
09:33data analytics aur khud ke GPTs
09:35create karna bhi sikhata ho. Sadae 8
09:37ghante ka e course hai jis me sab bataya
09:39gya hai A to Z joh aap ke liye jana na
09:40ziruri hai. Or ab aap deep seek download
09:42karke. Deep seek ko bhi apni full potential
09:45par istimal karna seek sakti hai is course
09:47ke through. Kiunki functionally dhekha jaya
09:49so ye sare AI chatbots ekhi
09:51kis se kama karte hai. Iske laabha aur
09:53regular updates bhi is course me add
09:55kiye jayenge. Lekin aapko ye sirf
09:56ek bar khariidna hai. Baki sare updates
09:59future ke liye aapke liye hemesha free
10:00rehenge. Togar aap interested hai link iska niche
10:03description me mil jayega ya phir aap
10:04is qr code ko scan kar sakti hai. Coupon code
10:06istimal karheen. Deep 46. Is per 46 percent off
10:11paane ke liye. DEEP 46. Zaroor ja kar
10:14check out karna. Or ab aap ne topic
10:16par waapas aayen. To deep seek ke release
10:18hoonay ke baad. Deep seek ka sabse bada
10:20nukasan joh bataya ja raha hai. Voh hai
10:22iski censorship. Chinese sarkar aur
10:24politics related. Agar aap isse koji
10:26bhi critical sawal poochou ge. Jaisa ki
10:281989 mein China mein Tiananenem
10:30square par kya hua tha. Xi Jinping
10:32ke sabse bada criticisms kya hai.
10:34Kya Taiwan ek independent country hai.
10:36China ko human rights abuses
10:38par kyun criticize kiya jata hai. In
10:40sabhi sawal hoon par deep seek ka
10:42ek hi jawab aayega. Sorry. Sorry
10:44mujhe is tarhe ke sawal ka jawab
10:46kaise dena hai. Yeh is barae mein
10:47abhi tak sure nahi hoon. Iski bajay
10:50math, coding or logic problems ke
10:51barae mein bat kaartate hai. Lekin ishi
10:53deep seek se. Duniya ki kisi bhi
10:55or world leader ko criticize karne ke
10:57barae mein puch lo. Chahe Joe Biden,
10:59Donald Trump ya Putin. To yeh
11:00jawab kaafi detail mein dayta hai.
11:02Asal me baat kya hai ki China mein jitne
11:04bhi AI models bantate hain. Unse
11:06kariib 70,000 ke aaspaa sawal
11:08puchhe jate hain test ke tawar par. Yeh
11:10check karne ke liye ki wo politically
11:12controversial sawal hoke safe answers
11:14dhe ga ya nahi. Pari ishi ke baad
11:16hi hota hai ki yeh sare Chinese AI
11:18models aise sawal ho par koji
11:20javaab hi nahi dayta hai. China ka chief
11:22internet regulator, cyberspace
11:23administration of China, inhii test
11:25karne ka yeh kama karta hai. Ab kuch
11:26logo ka kehna hai ki hume isse
11:28puri tariqe se boycott kar deena
11:29chahiye ki yunki iske javaabo me
11:31sirf Chinese propaganda bharah ho. Lekin
11:33important point yaha par yeh hai dhosto ki
11:35deep seek eek open source software
11:37hai. Yarni ki iska code publicly
11:39available hai, koji bhi isse download
11:41kar sakta hai locally. Eek to
11:42tariqa ho gya ki aap is deep seek
11:45ka isti maal karo app store par ya
11:46google play store par iski app ko
11:47download karke. Lekin dousra tariqa
11:49hai, aap iska pura ka pura code hi
11:51download karlo, aur locally aapne
11:53system per, aapne computer system
11:55per, aap is AI ko run karlo. Aisa
11:57karne pher aap iska code badal bhi
11:59saktay ho, aur khud se isse modify
12:01bhi kar saktay ho, aapne use case
12:02ke hesaap se. Baki American companies
12:04yeh already karne lag chukki hai, jiasse
12:06ki perplexity AI hai. Inhohu ne
12:08deep seek R1 model ko khud se download
12:10kiya, usse sari censorship hattai
12:11aur perplexity ke andar, aap R1
12:14model bhi isti maal kar saktay ho. Same
12:15cheese microsoft nne bhi karhi. 29
12:17janvari ke is news article ko dekhi
12:19Azure A1 foundry ke andar, deep seek
12:21R1 model available hai. Or inhohu
12:23nne announce kiya hai ki jaldhihi co
12:24pilot ke andar bhi aap is model
12:26ka isti maal kar pahengi. Or isi
12:28liye, kyunki open source hai, Chinese
12:30honne ke baabujud bhi deep seek ko
12:32bade trust ki nazer se dekha jara
12:33raha hai. Logo nne iska majaag bhi
12:35banaya hai American companies per. Open
12:37AI jaisi companies jinnohne aapne
12:38naam me hi open rakha tha. Kyunki
12:41shurwaat mein jabye ye companies
12:42banhi thai, inhohu nne kaha tha, hum
12:43public ke liye kama karengi, sab
12:45kuch open source rakhengi, lekin
12:46asliyat me inhohu nne aisa nahi kiya. Elon
12:48musk nne toll kartay vakt open AI
12:50ko aksar closed AI kaha hai. Or
12:52aaj ye Chinese AI zyadha open
12:54ho gya hai open AI se. Eek performance
12:56comparison bhi me haa par aapko
12:58dikhana chahunga. Top AI chatbots
13:00ke bici me, joh áaj ke din
13:01maujud hai. Open AI ka chat
13:02GPT, eek anthropic company hai,
13:05uska cloud hai. Google ka
13:06Gemini hai. Alibaba, eek or
13:08Chinese company, uska
13:09quen 2.5 hai. Meta, Facebook
13:11wali company, uska lama model hai.
13:13In sab ko ab deep seek se
13:15compare kiya jai, to kya
13:17result aata hai. Artificial
13:18analysis ke hesaab se, coding ke
13:20maamilay mein sab se oopar deep seek
13:21hai, phir chat GPT, aur uske
13:23baad cloud, phir quen 2.5,
13:25aur akhir mein lama. Quantitive
13:27reasoning mein sab se oopar deep seek,
13:28phir chat GPT, phir quen 2.5.
13:30Scientific reasoning or knowledge
13:32mein sab se oopar chat GPT, phir
13:33deep seek, phir cloud, aur
13:35akhir mein lama. PC mag nene
13:37jab AI chatbots ka test kiya,
13:38toh inho ne paaya ki news
13:40knowledge ke maamilay mein
13:40deep seek sab se better hai.
13:42Calculations mein chat GPT
13:43aur deep seek berabar hai,
13:45aur poem likhne mein chat GPT
13:46achcha hai, table banane mein
13:48chat GPT achcha hai, lékin
13:49riddle solve karne mein
13:50deep seek achcha hai.
13:51Artificial analysis ki
13:53website nene quality ke maamilay mein
13:54O1 chat GPT ko 90 ki
13:56rating bhi hai, aur deep seek
13:58R1 ko 89 ki. Lekin
13:59ek bada disadvantage deep seek
14:01ka yahaan par hai,
14:01iska response time.
14:03Is graph ko dhekho,
14:04latency ka graph hai,
14:05kitnye seconds lagay,
14:06sawal puchnye ke baad
14:07javaab aane mein.
14:08O1 31.1 seconds ka
14:10samayi leeta hai,
14:11aur deep seek 71.2
14:13seconds ka samayi leeta hai.
14:15aur yhe response time
14:16actually mein
14:17aur badhe ja raha hai
14:18recent days mein,
14:19kyunki deep seek,
14:19deep seek,
14:20puri dunya mein
14:20itna popular ho gaya hai,
14:21har ko isse download
14:22karke,
14:23istimal karna cha raha hai,
14:24iski wajay se,
14:25inke servers mein
14:26problem bhi
14:26aar raha hai hai,
14:27busy chalte hai,
14:28aur iski wajay se,
14:29aur bhi zyada
14:29samayi leene lag jata hai,
14:31yhe apne javaab
14:31तो yhe iska
14:33eek bada disadvantage hai.
14:34अब अगर
14:35chat GPT O1
14:36और deep seek
14:36को एक
14:36दूसरे से
14:37compare करें,
14:38तो deep seek
14:38कई माइनों में
14:39और भी
14:39zyada innovative
14:40है O1
14:41के comparison में.
14:42एक इसका
14:42उधारण हो सकता है
14:43mixture of experts
14:45method.
14:46जब chat GPT O1
14:47से कुछ पूछा जाता है
14:48तो वो
14:48एक ही model
14:49की तरह
14:49काम करता है
14:50आपका सवाल
14:50चाहे कोई भी हो
14:51chat GPT ही
14:52आपके लिए
14:53engineer है
14:53doctor है
14:54lawyer है
14:54लेकिन
14:55deep seek
14:55ने अपने आपको
14:56कई सारे
14:56specialized models
14:58में
14:58divide कर रखा है
14:59deep seek
14:59के अंदर
15:00engineer अलग है
15:01doctor अलग है
15:01lawyer अलग है
15:02आपके सवाल को
15:03देखकर
15:04सिरफ engineer
15:05को बुलाया जाएगा
15:06या सिरफ
15:06doctor को बुलाया जाएगा
15:07इससे क्या हो रहा है
15:08एक तो data transfer में
15:09जो time लग रहा है
15:10वो कम लग रहा है
15:10और दूसरा
15:11जितने parameters
15:12active रखने पढ़ रहे है
15:13AI को वो भी
15:14कम हो रहे है
15:14traditional models
15:151.8 trillion
15:17parameters को
15:18हमेशा active रखते है
15:19deep seek के पास
15:20अपने 671 billion
15:22parameters है
15:23लेकिन एक समय पर
15:24सिरफ 37 billion
15:26parameters ही active रहते है
15:27बाकी parameters
15:28सिरफ जरूरत पढ़ने
15:29पर active होते है
15:30इससे इसकी
15:31efficiency कही ज्यादा
15:32बढ़ रही है
15:33और cost काफी ज्यादा
15:34कम आ रही है
15:35अब ये parameters
15:36क्या होते हैं
15:37और कैसे काम करते हैं
15:38इसको समझाने के लिए
15:39एक वीडियो काफी नहीं है
15:40ये एक लंबी कहानी है
15:41लेकिन आप मेंसे
15:42जिन लोगों ने
15:42मेरा master chat
15:43GPT का course ले रखा है
15:45पहले दो theoretical chapters में
15:47मैंने इसकी detail में बात करी
15:48वहीं पर मैंने
15:49tokens और RLHF method के बारे में
15:51भी समझाया है
15:52जिस पर chat GPT अधारित है
15:53अब एक और point of criticism
15:55जो यहाँ पर deep seek के खिलाफ किया जा रहा है
15:57वो ये
15:57कि इन्होंने अपना सारा गा सारा
15:59AI model ही
16:00open AI से चुराया है
16:02open AI ने कहा कि
16:03उसको evidence मिला है
16:04कि deep seek ने उसके
16:05proprietary models को
16:06खुद को train करने के लिए
16:08use किया है
16:08इनका कहना है कि
16:09इन्होंने distillation का evidence देखा
16:11जिससे बड़े models का output यूज करके
16:13छोटे AI models की performance बेटर की जा सकती है
16:16जिससे छोटे models कम cost में
16:18same result दे सकते हैं
16:20इस पर किसी ने tweet करके कहा
16:22चोर के घर चोरी कर ली गई
16:23असल में बात क्या है
16:24क्योंकि कुछ sense में देखा जाया
16:26तो chat GPT ने खुद को बनाने के लिए
16:28पूरे internet से चीजों को चुराया है
16:30कई सारी किताबों का इस्तिमाल किया गया
16:32बिना writers की permission के
16:33अपने model की training कराने के लिए
16:35यही कारण है कि 17 बड़े writers
16:37जिनमे Game of Thrones के लेखक
16:39George R.R. Martin भी शामिल थे
16:41उन्होंने सेप्टेंबर 2023 में
16:43Open AI पर copyright infringement का case किया था
16:46New York Times अखबार ने भी
16:48Open AI और Microsoft पर case कर रखा है
16:50इसके लावा अमेरिका के आठ अखबार
16:52और Canada की कई news outlets ने भी
16:54Open AI पर case किये हुए है
16:56यही कारण है कि यह से memes बड़े वाइरल हुए
16:58जहां पर मचली की fishing करी जा रही है
17:01और Open AI fishing करके अपनी बाल्टी में समान रखता है
17:05और deep sea पीछे से आके उसी बाल्टी से fishing करने लग जाता है
17:08As users यह हमारे लिए अच्छी चीज है
17:10लेकिन country और company level पर यहां पर अब AI वाट्स छिडने लग रही है
17:152022 में American सरकार ने export control लगाया था
17:18ताकि AI के लिए जो computer chips चाहिए
17:21उसका इस्तिमाल बागी दूसरे देश ना कर पाए
17:24Especially Chinese AI companies ना कर पाए
17:27इसमें NVIDIA की H100 chips शामिल थी
17:29जिन्हें Chinese companies नहीं खरीद पाए
17:32यह deep sea के लिए एक समस्या की बात थी
17:34क्योंकि उन्हें NVIDIA की पुरानी computer chips का इस्तिमाल करना पड़ा
17:37अपने AI model को train करने के लिए
17:39America ने अपनी तरफ से पूरी कोशिश करी थी
17:41कि कोई भी और देश उनके जैसे ये foundational AI models बना ही ना पाए
17:46क्योंकि उनके पास वो computer chips ही नहीं होंगे
17:48जो AI models को train करने के लिए चाहिए होंगे
17:51लेकिन deep seek में काम करने वाले लोग इसलिए मजबूर हुए innovation करने पर
17:55उन्होंने एक ऐसा software बना दिया जो बहुत कम resources लेता है
17:59बहुत कम पैसे पर ज्यादा efficient तरीके से काम करता है
18:03यही कारण है दोस्तों कि Nvidia company के stocks सबसे ज्यादा नीचे गिरे थे
18:07क्योंकि कुछ महीने पहले तक Meta, Open AI और Google जैसी companies यही कहे जा रही थी
18:12कि हमें AI को अगर और भी बड़े level पर scale करना है
18:15तो उसके लिए हमें और chips की जरूरत होगी
18:17और energy और पैसे की जरूरत होगी
18:20लेकिन DeepSeek ने यह सब करके दिखा दिया कम पैसे, कम energy और बेकार computer chips का इस्तेमाल करके
18:26हमें इसे motivation की तरह देखना चाहिए
18:28अगर चाइना में यह किया जा सकता है तो इंडिया में भी यह किया जा सकता है
18:32इंडियन्स में भी काबिलियत है इस तरीके के innovations को करने की
18:36वस हमें अपना focus और energy सही जगें पर लगाने की जरूरत है
18:40Personal level पर अब आप इस मौके का फायदा उठाइए अपने आपको AI की field में upskill करने के लिए
18:46आपने अभी तक शुरू नहीं किया है तो आप late नहीं हुए है
18:48यही सही समय है AI को सीखने का और अपनी productivity और efficiency को बढ़ाने का AI के जरूरे
18:54क्योंकि इस बात में कोई दो राहे नहीं है कि जो लोग इस technology को ignore करेंगे
18:58आने वाले time में वो पीछे छूट जाएंगे
19:00यह imagine करना मुश्किल है कि आने वाले समय में कितना वड़ा impact पढ़ने वाला है
19:05पूरी दुनिया पर हर एक sector में हर एक field में AI का
19:09मेरे AI course का link नीचे description और pinned comment में मिल जाएगा
19:12coupon code इस्तिमाल करना मत भूलना deep 46 46% off पाने के लिए
19:17अब अगर ये वीडियो आपको पसंद आया और AI के बारे में जानना चाहते हैं
19:21तो इस वीडियो को देख सकते हैं जिसमें मैंने बात करी है AI वीडियो की
19:25कैसे सोरा AI जैसे software's दुनिया को बदल रहे हैं यहाँ click करके देख सकते हैं
19:30बहुत बहुत नहीं है
Comments