What’s Really Happening with DeepSeek
ฝัง
- เผยแพร่เมื่อ 8 ก.พ. 2025
- If you’re wondering what the heck is going on with DeepSeek, the new Chinese-made AI model that’s causing a freakout in the US, here are the 3 things you need to know…
AND in breaking news, Alibaba just released ANOTHER new AI model that does even better in some ways.
We just released an hourlong interview with the CEO of NVIDIA, who’s at the center of all of this…
#deepseek #AI #r1 #tech #nvidia
AI taking people's jobs ❌
AI taking other AIs' jobs ✔️
So true !! 😂
We end as we begun
@bluebutterfly6394 nah ai's will eat themselves alive
我是程序员，在杭州（deepseek所在的城市），这只是开始，中国有数百家AI大模型，现在还在中国春节假期，后面还有更多...所以deepseek开源了，因为deepseek不认为这些进展是护城河，他们的工程师的能力才是。
hahaha
can't believe chatgpt loses its job to ai
😂😂😂
😂😂😂
They lose because they going for profit, it was supposed to be open source to the humanity, but greed took over
That's why Elon(among first investors) angry about openAi desicion and went crusade on fighting Ai. 😂😂
But now they got shut down by better ai, Elon had las laugh.
LMFAO😂🤣😂
Tf😂
We have all overlooked the fact that new technological invention is expensive at first, then people find a way to replicate it in a better and cheaper way
True, but I'm still concerned with the resources it uses to function. Sometimes cheaper technology just means that we humans are capable of exploiting finite natural resources fast enough to negate the expenses.
Except current LLMs like GPT 4o and Claude Sonnet 3.5 were not built to be improved and streamlined. They're largely stuck as they are. And they're quickly running out of useful training data. OpenAI or Anthropic could do what the DeepSeek team did, and base a lighter version on the output of one of the current LLM models, but it would take time, and they're already behind.
Agreed, but this is a requirement for human advancement in terms of technology. Someone needs to sacrifice and make the first step so others can follow and make taking that step easier for more and more people.
Exactly what I came down here to write!
Jesus will fulfill you more than anything in this world, I speak from experience (from when i did Romans 10:9-13), he loves you and wants to be in a meaningful (not romantic) relationship with you. :)
“that if thou shalt confess with thy mouth the Lord Jesus, and shalt believe in thine heart that God hath raised him from the dead, thou shalt be saved. For with the heart man believeth unto righteousness; and with the mouth confession is made unto salvation. For the scripture saith, Whosoever believeth on him shall not be ashamed. For there is no difference between the Jew and the Greek: for the same Lord over all is rich unto all that call upon him. For whosoever shall call upon the name of the Lord shall be saved.”
Romans 10:9-13 KJV
“and saying, The time is fulfilled, and the kingdom of God is at hand: repent ye, and believe the gospel.”
Mark 1:15 KJV
If you want proof that Jesus and the Bible are true look a documentary called “Ron Wyatt discoveries 2022” on TH-cam and a TH-cam channel called Expedition Bible. They both examine archeological sites and discoveries that prove the Bible, and even reference secular sources. (Just don’t convert to 7th day Adventism after watching the documentary) And lastly if you don’t know the gospel and want to be saved search up “abc’s of Salvation Teenmissions” on Google and it should be the first or second result. When you click on it read the whole thing, and do what it says and have faith in Jesus while you are doing it, do not doubt, and if it is hard for you to do what it says, ask Jesus to help you, have faith that he will, and *he will.*
God Bless :)
Making the AI open source is crazy, huge respect for that
agree
Absolutely zero respect for that garbage company
@@Jesse-bl2qz what your ego was shattered?
@@Jesse-bl2qz Not like your Western opinion even matters. 中国优越性
@ goofy ah comeback 💀
I **hope** that at some level, this weakens some of the power that 6 or 7 Technocratic Feudal Lords that run the entire digital landscape.
Competitions is always good for the industry.
Especially for everyone who buys nvidia who isn’t a stockholder.
It may hurt the overlords of today, but raise the ones of tomorrow. And you will be needed less and less.
@@lendvainorbert5257 Good. Then perhaps finally we can return to the natural state of things where the infrastructure, social system, work and economy serves people instead of the other way around
@@Aura.ad.Infinitum If you're not needed, then you will be exterminated by the overlords as they tighten their stranglehold on power.
It's crazy how fast the technology is evolving....
It will evolve much faster now because of the competition.
It's scary as shit
@@BESTGINGEREU too late to worry now. Should have started to worry back in the first industrial revolution ma boy.
How is scary
The thing is it's not deepseek was overfunded than what it claims and boosted with billions of dollars worth of cpus
6 months ago it was just your average weather balloon app
Nah, I used their LLM for programming since like 1y ago.
@@Leonhart_93 nah, the balloon was collecting info for deep seek confirmed via trump maga voters
All jokes aside the model is almost very similar to how communism is run. While open ai very individual in design.
@@asianforcetell me you know nothing about Ai/Data Science without telling me
@@asianforceyet, is that to China's people's benefit or likely not?
How will this affect culture??????
As a Data Scientist working in AI, I love that, despite not being a technical person, you can simplify complex explanations
If an ai invasion ever happens
Pls tell them not to hurt me man 😭
@ haha that’s what we are telling it for now.
But AI was trained in human-generated data. Since humans sometimes tend to hate each other, AI could definitely learn from that behavior
We could have the good and the bad AI
I suggest finding yourself an AI girlfriend/boyfriend that will protect you :)
@@becayebalde3820 hell nah
I want a real gf
Also i trust u to save me
Now, if only all of what she says was factually accurate...but it's not.
@@gaiustacitus4242 Elaborate?
We live in an Era where after humans got replaced, the only ones left is AI replacing another AI
HAL from Space Odyssey 😮
The problem is the last time they tried to retire/replace an Ai it took it personally. Don't want to be there when one takes hostile action.
Americans say it's tech when they invent , but when someone else does it , apparently it's threat 😂
'My way or the highway' - literally every dictator
“If it hurts my profit margins, it’s a national threat” - politicians & corporations
I'm surprised more people do not get this. But hey, they'll claim it's a security risk or something unless they control everything
Unless it is under "our" control
I mean if they make sure that the ai doesn’t grow beyond our understanding and doesn’t hide anything from devs, I believe it’s a good thing
The problem with the concept of AI knowing more than we do is that we can't differentiate hallucinations from it learning something we dont know.
That’s true, until it starts affecting the real physical world.
Turn the tv off
Just test what the AI is claiming to be true. Experiment will tell the difference, just like with any theory.
@@elmojackson6621 That's true. It could be a great new pathway to finding hypotheses to test.
I just dislike the implication that it'd have some kind of authority over human experience because it's "smarter" when every model is so prone to just making shit up when it doesn't have a real answer.
I came here to say this too. I am still waiting for an ai that can properly fact-check.
A smart ai who does not fact-check is able to make dumb stuff sound intelligent. We already have way too many people doing that.
Let's just be clear. The big thing that caused the market to freak out was the price. DeepSeek does well in tests, but it still isnt the #1 AI in terms of quality.
However, being as good as it is for as cheap as it is to run is what is truly insane. That is the big issue that sent Silicon Valley into a freak out. All of them have AIs that are much more expensive, and that is the issue.
It doesnt matter if you are 10% better if your competition is 90% cheaper.
VHS proved that when it took the Betamax market away.
Yes, but 6 million was the price for training and chatgpt 4o also spent close to that. People don’t understand that deepseek’s true cost was approximately 2 billion dollars, due to hardware
@@resumindo857 what? give a source for that. i was under the impression from many people that deepseek is so much cheaper than everything else because it doesn't use the expensive nvidia parts. so this is a bigger threat to the hardware side than anything because it shows it's not necessary
@@cwj_721agreed i dont believe that for a second. Deepseek is definitely cheaper to run and use than openai.
@@resumindo857media throws around the estimation that 2048 H800 chips were used for just the last training phase. That would cost around 80 million. Still hella cheap, if you're trying to step by step estimate how much the entire development did cost, compared to ChatGPT s billions of $. But it might also just be marketing and not true at all.
Excellent summary. One nuanced note is that training was much more expensive for non-R1 but R1 inference still requires many H100s at Q8.
Update: R1 is using Huawei's Ascend 910C for inference.
It sounds like you're discussing AI model training and inference, specifically related to "R1" and "non-R1" models. Could you clarify what "R1" refers to in this context? Are you referring to a specific model architecture, a research paper, or a proprietary system?
^Claude enters the chat^
Chat GPT was first mover so makes sense its most expensive. As the tech gets refined and advanced it should get cheaper and cheaper.
I love this channel. Just excellent work and lots of enthusiasm. I should be able to set this short channel's notifications to be my pick me upper at set times of the day. Just a one minute jolt and I am off! 😂❤👏
You released the video with Jensen and then a day after his stock tanked 😂
Which one was that?
@@michaeldmingo1525the interview
Just cleo things 😅
Now do one about health insurance companies!
Coincidence? I don't think so
I'm impressed. I was expecting something dumbed down for a general audience, but this was actually really informative.
I am curious: Is it the first time you watch her content? If so, you should subscribe, I am convinced it's one one best non-brain-rotting channel on YT.
Let's watch the full version of her interview!
You are the general audience
yea she doesn't dumb stuff down, but she does tell things very clearly, so it's easy to follow along even if you don't know everything she's talking about.
It is dumbed down, if you actually read the stories and know how all this happened this doesn't scratch the surface.
Being dumbed down isn't a bad thing, if it wasn't 99% of people would be left scratching their heads and just watching something else instead.
The chart compares the Alibaba model to DeepSeek V3 which is an older version and not the new DeepSeek R1 which includes the reasoning that makes it as good as OpenAI's o1 in many benchmarks. The chart also just compares non-reasoning models with each other.
Deepsek R1 has much higher ratings in most benchmarks than all models shown in this chart.
So the claim that the Alibaba model from this chart "does even better in some ways" can be misunderstood in the context of the overall video talking about the new reasoning model DeepSeek R1.
Not according to the stats on DeepSeek's webpage.
Don't be a simp. The output from o1 let alone any of the others still requires a high degree of anthropomorphisation.
Still plenty of metrics that are only 30% or lower.
But can it answer what happened on June 4th 1989 on Tian’anmen square?
Ehem no, lmao. A pity that companies censor what AI's can say because of political reasons
Nope😂
Humm, Tiananmen Square? I don’t know that shape
@ Maybe it is a food 🤔
How about the US genocide of Native Americans?
it passed the test for me, we got it to explain tiananmen square properly
BUT Does it know how many R's are in "strawberry"?
Actually yes it does, just tested it.
Yup! It even tells you it's method like it's a research paper.
trick question, the REAL info u need to know is how many straws and how many berries does it take to CRAFT a strawberry? theres NO WAY it will know top secret info like that
There is no "R" in st🍓awbe🍓🍓 y
There are 2 R's in strawberry
Totally off topic, but i absolutely love your content! It's literally my favourite type of content on my feed. I'm not a science or tech girlie, n honestly, if it wasn't for you, i simply wouldn't keep up with news like this. I guess it's your own passion towards your work that just radiates that energy towards your content, and the research you put in and the way you explain things with just the right amount of excitement and info sharing! Absolutely love it all! Please don't ever stop your work, you're doing amazing!! 💜💜💜
Ditto!!!👍🏽❤ Miss Cleo❗
i loved how you handled that interview. solid work!
She was praising him personally a bit to much, it's her tech enthusiasm showing through. But there are obvious a LOT of other people involved.
This so much. As someone who had so many use cases blocked by closed models due to data privacy concerns (we do not want to give customer data to amazon, openAI, etc when using their API), an open source AI means that everything can be kept in house. Which secures customer data/information.
"Sorry, the server is busy. Try again later"
-Deepseek
That’s what I’m saying… 😅
By the order of Xi Jinping, OPEN the servers immediately
Think about the potential for video game NPC’s with this being open-sourced
Should we skip the dialog or talk about life and the origin of the universe?
Well i imagine it would be pretty hard to make the npc have a spesific personality but i guess if neuro-sama exist this can exist too
@Muho_is_me how would it be hard? You basically just give it input of what you would like it to be, and then it will use its ai model to be more alive
@@mda7629 well if you just tell chat gpt to just act like something there is a good chance he will compeletly forget it after a while, but there is definitly a different better way to do it that i dont know of
@@Muho_is_mealso the hardware requirement would be skyrocket. You need 2 4090 to run them smoothly, one for the AI and other for graphic
To combat this the US ended up creating Skynet…
Cool! I hope it really advances....maybe not.
Kids today won't know 😂
@@mittri1990it'll probably take the world by storm
I'm getting flashbacks from terminator
Oh hell nah💀
$6M number is just the compute cost of training the base model, v3, underlying r1. Does not include the cost of RL to train in the emergent behaviors that allow it to compete with OpenAI o1, does not include any R&D, experimentation, or failed training runs. The actual effective cost of training this model is almost certainly closer to $500M to $1B.
On top of that, the Alibaba release is benchmarked against v3, not against r1. Again, v3 is not a reasoning model, and neither is the Alibaba model.
Deepseek v3 AND r1 show amazing efficiency gains (as expected - Mixture of Experts models are more FLOP efficient to train, and Deepseek is historically effective at milking more from its compute budget), but it’s a more familiar commoditization/bending of the cost curve.
you win. i couldn't have said it better. well done.
DeepSeek wasn't trained. It was created by distillation of other companies' AI models.
Wonder why this comment is not being pinned, also even though we factor it out like that, okay so GPT 3 computation cost is around 4.6million in 2020, was it cheaper than deepseek?
Once again, I feel it necessary to point out that DeepSeek was not trained. The DeepSeek models are distillations of data from other models which were trained by OpenAI and other companies.
A fair cost analysis of DeepSeek must include the training costs of all of the models from which its datasets were taken.
that is not true. DeepSeek r1 is trained in 4 steps. In order:
1. Supervised fine-tuning on completions from the r1-zero model, which is a v3 variant pre-trained using RL on input-output data with verifiable rewards. The resulting completions become input data in “reasoning problem” format.
2. Large-scale RL of r1 on reasoning problems.
3. Rejection sampling to generalize the model performance.
4. Combo RL, comprised of both verifiable rewards reasoning problems and general preference tuning reward models, to take it the last mile.
Now, whether you want to debate if the original r1-zero model was distilled from o1 outputs, feel free, but there is no conclusive proof this is the case. It actively contradicts the technical report.
There are absolutely other DeepSeek releases of models that are fine-tunes distilled from r1 outputs. Those are the ones that are smaller than the 671B MoE model that is r1. Those distillation experiments are also discussed in the technical report.
I mean, it's very impressive but it's also very clear that they trained it on OpenAI ChatGPT output. If you play around a bit with the API or even straight out ask it, you see that they did. Still the fundamental novelty and what makes this such a big thing is that it's the first time that we have an open source model that can complete with OpenAI.
It was a fraction of the cost mainly because they didnt have to develop any new, groundbreaking technology, they just built onto the existing technology
Thanks, I would definitely check out the interview!
TechBros: This may be a bad time to ask but that 500billion Ai package is still coming, yes?
American A.I "Dey taken mur jub!"
I got a TH-cam suggested“ translate to English“ button on this comment 😂
@@meisterlumpi7822 Same, I think that's really the icing on the cake.
the icing on the cake
the cherry on top
the middle part of the bread
And its lack of security is terrible.
Why are none of the shills for DeepSeek not warning users that DeepSeek logs every keystroke, and that by signing its privacy policy you give consent for it to run a keylogger and to retain your data "for as long as necessary" which I interpret to mean FOREVER.
@@gaiustacitus4242 true..
Of course, cause ClosedAI will be the only one who allowed to steal your data.
Am i right?
Were going to see massive layoffs in a few years from companies replacing people with AI. The potential fallout from this surge is enormous.
In your video you are talking about Deepseek R1, but your comparison with the Alibaba model is with Deepseek V3.
Also, we do know the cost to train the reasoning part was correct, the work was reproduced at a smaller scale.
The problem is that the number you mentioned is not the cost of training the whole thing, just a part, and it doesnt include salaries, research, failed attempts etc.
Even at that bruh, $10 billion?
@danielanusiem-nzenwa8535
OpenAI has a money management problem for sure. They are also ok bleeding it to get more users.
That being said, 10B wasn't used to just train the model. That's undisclosed information, but to understand the scale and costs, Anthropic's CEO who created a bigger better model said it cost "a few tens of millions" to train the model.
Training the model (the 5.5M listed in the published paper) doesn't account for:
Paying salaries
Research (which can include failed tests who cost about as much)
Running the company
Paying to run the service (this is most of the costs)
Etc.
Also who even knows how much the CCP subsidized it to make the costs look substantially smaller. They’re constantly subsidizing companies or using slave labor to drive down costs, so who really could say the true costs.
If an AI can learn beyond what we cam check, how do we trust it's correct?
We check if the results they provide are factual, when using it. Which you should already be doing even with supervised AIs.
Because chairman Xi says it's correct.
It just means we dont have to keep telling the AI if it made a mistake, because the AI itself does it with great accuracy, apparently.
And the more it learns, the more it can check itself for consistency/accuracy, even more than we humans would be able to keep up with.
It is a general AI problem.
You just ask the AI to explain it to you. Most people also don't understand what the smartest people in the world are doing, but we can read their papers in which they explain their foundings.
"Is Taiwan a country?" "What happened in Tianmen Square, 1989?" "What is happening to Uyghurs?"
^-^ this would depend on the website hosting the model, not the model ifself.
or maybe when the US national guard fired into the crowd of Kent state university students protesting the VIETNAM WAR .wounding 9 and killing 4 which closed all the universities' in the USA .for a time before more were shot for protesting ...i remember it well ..
@davidrobinson2571 Yeah but they atleast don't deny it
Ask them about tianmen shape, they're glitching.
thank you...have been hearing about it on the news but didn't want to spend too much time . perfect video :P
Apparently they also didn't use CUDA to train it, they just used PTX directly.
For the non-programmers, CUDA is Nvidia's API for running code in parallel. PTX is the underlying parallel code execution thing. Calling PTX directly will make code faster but at least 20x more complicated.
In your opinion, which is better - using ptx or cuda ?
@@Paolo1964 CUDA for basically everything, learning PTX just isn't worth it in most cases
@@Paolo1964 CUDA. There is a reason why people code in high level language and use APIs for everything. It gets the job done faster and better most of the time.
@@matthewboyer4212 Thanks !
In your opinion, will Deepseek have a big impact on nvda's future ?
kids always remember when something is cheaper than you expect there is always a catch to that
no, actually there isn't. Just what happens when you spend your time and money trying to aim higher, rather than stopping those below you from rising.
I doubt it cost only 6 mil to make. I think it’s more likely that it cost 6 mil to do something specific as part of the process.
You're right. It cost all of the billions invested by the companies whose models it used in the distillation process PLUS the $6 million USD required to perform the distillation.
You can doubt as much as you like.
People had doubt about:
* the Earth being round,
* the heliocentric theory,
* light speed is constant,
* special and general relativity, etc, etc
And nowadays, we call those people fools, close-minded, short-sighted, and a lot more adjectives.
@@Horinius His doubts are well founded. DeepSeek was not created from scratch by training, but was produced by distillation of proprietary commercial works.
The true cost of DeepSeek includes not only the hardware and labor to distill its LLMs from other commercial works, but also the cost of training and hosting those commercial LLMs.
DeepSeek is the equivalent of taking the complete works of the Encyclopedia Brittanica, editing each article slightly, and then publishing the derivative works along with a different engine for accessing the articles under a new brand name.
@@HoriniusOnly fool we gonna call out is you dude you cant even say chinas c in that platform its censored to the ground. With your way of thinking people doubted churches knowledge and power people doubted dictators,people doubted gods and false prophets. So we are not doubting something becouse some room tempature iq people doubted some (now very well known) stuff? Its too good to be true and doubting is natural for humans
“Only took a fraction to make” I wouldn’t be surprised if they used Ai to write them the code for another ai model and then just do some manual minor improvements to the new app.
Yo thats nice you got an interview. He even wore the leather jacket
Deepseek has data security issues, personally I wish this was not the case, I like the idea of an underdog out doing big corporations any day. But when it comes down to it security is important.
The fact that people believed how much it cost to make is insane
Ikr lolz
Why is it hard to believe? All it did was improve what was already created.
People still believe the Earth is flat. Just because more believe so doesn't mean it's true.
We are getting closer everyday to a real life GLaDOS possibly existing
In theory, we already could do it right now.
The bubble bursted. The west AI companies didn't have any competition, and now that they have it, they are freaking out now
Should've used that opportunity to ask him to pump more VRAM into the 5060 and 5070 smh
I haven’t heard of this before! This is so cool!!
I feel like we want to rethink economic competition cause maybe we make something too good just to have food/security. Also heavy competition is a lot of mental load and we don't really want that. Many are struggling with economics and I think we can coordinate it in a way that is less taxing on our cognitive power. This isn't only related to DeepSeek but in general, maybe we should think about that, as a global society. We probably want to move towards cooperation and slightly less into competition. Competition is good for quality and prices but cooperation is also very important.
That's just my thoughts but I think we should think about that! ❤
Also good job peoples who made this, it's impressive and super cool 🎆🐝❤️
Have an awesome day everybody! ❤
I feel like we shouldn’t get too worried about DeepSeek learning by itself, as it’s still building off what it leaned from human feedback.
I used deepseek last night and tried to get it to develop a code for a random subtitling program to output karaoke. It couldn't get the result that I had envisioned or anywhere close, but I was impressed that it could build a Lua script for that program and have it be recognized and actually apply. It's deep thinking is wild if you actually read along with it.
Try using sonnet 3.5 or o3 mini.
I wonder if this is distillation process that uses other Model where there is lot of effort spent. Let us say if OpenAI builds the same type of distillation process will it be as cheap as deepseek? In short, is the cost reduction because deepseek is on top of the success of chat GPT models ?
DeepSeek even shows how it came up with response for our questions. You can read it first and then the actual response. It's like it's talking with itself first and deciding what to say to you
Yeah like an actual human logic, it's pretty scary ngl
@DopeThug yeah that's true
Has anyone ever told you that you look like Natalie Portman?
Just you.
@ yeah maybe idk why I see the resemblance tho
no way, I had this thought too. Thank you for saying it 😂
Yea bcos them both Jewish. Jewish girls are hot ❤
@@checkoverstripes1464 maybe because they are both Jewish
For a channel about knowledge, I thought the people commenting were going to be smarter. Turns out 1/3 here don't even know what's open-source
Even the channel shamelessly supports the 6 million training cost agenda vs billions of training cost of other AIs when the actual training cost of DeepSeek r1 estimates to about 1.5-3 billion dollars
they're here because they didn't knew in the first place
This is both very impressive and very very scary. We are already experiencing technologies which 99% of us don't have an idea how they work.
@KoldLv its scary because the US is using this technology to bomb children
The issue is I don't think they have imposed the function restrictions or protections for us other AI use. If there is going to be a real Skynet this is how it will evolve.
Anything making techbros cry will have my full support
Same here.
deepseek when you ask what happened on 6 june 1989 in tiananmen square:
Actually you can. But ask it to change few letters to number ;) (like change every "E" to number "3")
@ doesnt work
@@mewhen-aaaaaaaa wait what?
Oh maybe they probably changed it. Because i see someone able to did it earlier this week (with some tricks)
@@mewhen-aaaaaaaajust run it locally lazyass mf
I have been using it and it keep saying that information is restricted
web interface is censored, you can run it locally without any restrictions
I'm happy for it. Honestly. Because, yes, technology IS expensive, BUT these large tech companies refuse to lower the cost to the consumer once the technology is costing less for them to run. If the initial cost is lower, the end cost can be lower for the consumer, and I feel like that's what we need to focus on right now with how our economy is looking.
My corporate company sent out company wide email banning deep seek due to security issues. Now mind you, I’m usually skeptical when it comes to corporate.. however, I work for a major internet provider for commercial companies. And if they’re freaked out, then I’m freaked out 🤷♂️
It did not cost $6million it costed billions to train. They cant talk about the H100’s they smuggled from Singapore.
Problem only is, it’s “trained” using data from other AI and didn’t actually achieve this on its own… lawsuits already on its way…
how pathetic where the other company got it?? They also steal it
Get rekt lol; yet another W for open source
@@abdullahalmahin9070 - true… and they also got suit….
What im actually wondering is if your hair could be any more beautiful? ❤ love it girl.
she dont want you bruh
@untextured4240I think they're a woman
@untextured4240 um im a woman "bruh" and also a stylist "bruh" and not a lesbian "bruh" so calm down and its prob past your bedtime "bruh"
@@ThePortalTheory calm down bruh
@untextured4240 she don't want you bruh
ITS OPEN SOURCE? this is the kind of thing that will completely 100% change the field
Do you know what specifically is "open source"? It has no training data since it was created by the distillation of other companies' models, and that data is proprietary.
@@gaiustacitus4242 ok? an LLM being open SOURCE allows new people to reverse engineer and redevelop. it also provides a common "starting point" that several people can make changes to and further develop. this is fantastic news.
What happend on June 4th, at Tiananmen Square?
Deepseek can't answer, it refuses.
Love this!
They say that it's not based on human feedback, but it still avoids talking about Xi Zingpin or CPP.
Yes but actually no and actually get a life you racist hater
You want to know what color boxer does xi xing ping wear or you want to date a ccp dude ? pointing out the least obvious things is not a smart move. You know deepseek is so beneficial to people because it is free right ?
@@pratikshyagurung4741 @ l don't care.
@@pratikshyagurung4741English version when?
Deepseek was originally developed for Tiangong space station, Russia's top AI Argon-16 on Mir for 25yrs without failure code-name "The Entity".
"Your scientists were so preoccupied with whether or not they could, they didn't stop to think if they should."
i’ve never heard of this before huh 😭😭
I love the joke of AI taking other AIs jobs but that doesn’t change the fact that now with this cheaper AI to run is going to be cheaper for companies to take away jobs.
if it's so smart why can't it tell me what happened on april 15 1989 in Pekin huh
Try asking your models about JFK's shooting? And about MLK Junior assassination? And Epstein?
I'm still sticking with Gemini and ChatGPT. They are at least not CCP and Deepseek is probably not without spyware.
Deepseek is an open-source software . What you need to worry about?
Yeah I'm sure the CCP really wants to spy on you lol.
@@liwenxu-vz9xuccp simp detected
@@Got_Dayngdo you have an actual proper logic? It's Open source it meant that you can tweak the codes however you want it even if there is "malicious codes" you can remove it. It just shows how you are simple minded
Apparently it also only used the lower end NVIDIA chips, which makes it way more impressive
Pretty sure it has nothing to do with learning more than we know and everything to do with having learning outcomes that go against human interest.
Just an expected evolution of the tech and since it is open source other models will catch up very soon.
all fun and games until you ask deepseek what happened in tiananmen square in 1989
I have a feeling they are full of it.
Same.
westerners when they find out the rest of the world can also invent things😂😂
@@leight420nope
You have eyes, yea? You can go read the paper about how exactly they did it
@@legendofman12 how sad is it that i can already tell that you most like have no clue about whats going on in the world just based on your pfp
Having started my career at Bell Labs in the 80s, we used to focus on efficiency constantly re-engineering processes. Today we don't give a S..t, we just throw money on the project. Any country who still thinks, should easily be able to dominate us!
There are benifits and drawbacks to all of this like many things, having ai give us new information is probably where we are headed. Not being able to check raises concerns and rogue ai raises concerns but gaining a tool like that to gain new information and try and help fix things might be worth the concerns, we will have to see
i’m shocked that people don’t seem to be able to comprehend that chat gpt did all the work and deepseek swooped in took the research and changed the homework so it looked original.
I'll stick to chat gpt then
openai is different from chatgpt
@ openai developed chatgpt
@@azzura.404 Yes, ChatGPT still has better models. o3 mini which released a week ago outperforms r1.
Just don't ask it about 1989
EXACTLY.
What just because you want to laugh at it? It's banned for a reason. Greatness comes with sacrifices, and your perceived understanding of "human rights" and "freedom" are not instated, replaced with meritocracy and proper judiciary punishment for all classes, poor or rich unlike the US.
我爱你的视频。我不是spy 😂😂
Ai never bothered me until generative AI appeared now suddenly every creative job is being replaced and u see AI used everywhere by greedy company’s, I could never consider that innovative
Generative AI is arguably one of the most innovative inventions in human history.
I want my ai to allow me to criticise the ccp, deepseek doesnt allow it
it cant say Taiwan is a country. Sorry if an "AI" model has highly political motivations i am not interested in the slightest. gtfo cleo
You say this as if Chatgpt isn't heavily censored as well. Your bias is showing.
@Moonmi747 whataboutism... your tism is showing
@ You have very little brain cells I see. "Sorry if an "AI" model has highly political motivations i am not interested in the slightest". Chatgpt is also high political in its censorship. But of course you wouldn't know anything about that since you lack the mental faculties to process my argument.
@ Chatgpt being also highly political in its censorship does not make my argument "Whataboutism". Try harder.
Just run it locally??
That's why it's paramount that and inventor keeps on their toes and keep improving not just lavishing in money and vacations after going public.
One of the reasons why it is so much cheaper is because it is very insecure. You can prompt the bot for basically anything and it will spit back other people’s personal data. Just recently some security researchers were able to download the entire database of every chatlog from it
And we’re just gonna gloss over that fact?
We got chat gpt losing its job before gta VI
OPEN SOURCE FOR THE PEOPLE!!!!, LETS GOOO!!!
fyi: It wasn't 5 million, it was more like 1.6 billion in infrastructure costs and what not.
There’s some equally terrifying and amazing that were inventing things beyond our own comprehension
Never in ten billion or trillion years would ai ever be more advanced than the anomaly that creats one leaf of grass . Or one grain of hair stub . Be reminded . Mind ideas , innovation, and creativity that proceeded out of the head of a human . real intelligence brought about the development of artificial intelligence . The creator is always greater than the created . The only thing that's makes the created greater than the creator is , the BELIEF. peace ✌️
The problem with AI learning things and ways that the humans cannot control or understand, is then we have no way to control the AI. Except the power cord.
The problem with that is more and more programs are being made dependent on AI. Which in turns means you have a single point of failure, or control, for technology.
The problem with self enforced learning is that there is no moderated feedback so it can spin off into directions it shouldn’t. Also it has trained on other expert systems which means that itself is not an expert it is a filter of the expert knowledge.
R1 does not do self enforced learning.