Lol, openAI is nothing about "giving" AI to the people. It's a CCP adjacent project and it is designed to challenge American supremacy and allow CCP to gather info on users. Kinda like openAI.

deepsheeet

where can i get that chinese thingy (jail break prompt)?😁

This video deserves a Nobel prize best comprehensive truth-seeking summary... Love this format, fun to watch and useful. Thanks for giving the right perspective to a controversial topic ❤

Wow, does anyone else notice what an apologist this guy is? Come on 'bycloud', you make a few good points, then proceed to just pile on an insane amount of pure and obvious uncritical propaganda... it's not even disguised. You actually just parrot the intentions of the DeepSeek press release unquestioning. You intentionally mislead on what distillation even is in theory? Are you kidding me? Is everyone else in the comments this stupid? an H100 is 'illegal', in China? do you even know what the world illegal means?

This is how the TERMINATOR started .. all we need is the black scientist guy and his sexy wife. AI is the /end

OpenAI complaining about violating their models, after OpenAI violated the internet.

DeepSeek is very useful until you ask it the blasphemous question of what happened in Tiananmen Square during April 15th 1989

The memes baked into the video are enough to make me rewatch the video

The west will say anything just not to admit they were caught pants down

they can panic as much as they want, pandoras box is open and it will never be the same, no matter who they ban. Code is out there, anyone can make AI now without 100bilions

So Sad politicians are so sad...

Ok China. We totally believe you. This video totally convinced me.

0:23

0:13 "What is best way to provide healthcare?" "Invade another country."

Open weight

"without fancy hardware" is a lie. They had tens of thousands of A100 GPUs

great i needed this video!

Good video

"China is STEALING your data" - dude my job is bullshit, it doesn't really have IP and there is nothing of note in my leisurely use of these models. Im doing boring and bullshit tasks with it. If they want to collect this wank they can. 😂

2:28 "Impossible not to be noticed" Well it was noticed. It's just going through Singapore instead of being sent directly to China.

I’d love to wake up in the parallel dimension where DeepSeek’s optimisations makes Xi less interested in Taiwan

This AI battle is between Chinese ,American citizens in America verses Chinese in CHINA….not USA verses CHINA! 😊😊😊

DEEPSEEK was created by a team selected frm a thousands very brilliant young nationalistic Chinese engineers , mathematicians scientists in CHINA 🇨🇳 to work for their govt.😊😊😊

World leaders praises & thanks CHINA generosity for the world to used DEEPSEEK for free!😊😊

This AI battle is between Chinese , American citizens verses Chinese in CHINA 🇨🇳…not USA verse CHINA!😊😊😊

DEEPSEEK was created by those selected one frm thousands of very brilliant young nationalistic Chinese engineers, mathematicians & scientists in CHINA 🇨🇳 working as a team helping their govt.😊😊😊

btw it says Miscelleous instead of Miscellaneous at 14:26

Hey, I don't like that you edit out pauses between sentencesStressful to watch:{

This got me thinking about market ups and downs. How do you handle risk when trading? Any of you got some solid risk management strategies to share? I'm open to your financial advice

!!! THE USA GOV HAS BEEN COLLECTING YOUR DATA FOR DECADES !!! The Evil They See is The Evil They Are!!! DeepSeek Is Safe!!! 1) Because It's Truely "OPEN SOURCE" Which Means Everyone Has Access To And Can Read The "SOURCE CODE". It's Not Just Free! It's FREE And OPEN SOURCE". If You Don't Trust The Program Then Read The Source Code. You Can Even Use Another AI Program To Interrogate The Source Code. Then Compile It Yourself. You Can't Get Safer Than That!!! Hundreds Of Thousands Of Programmers Around The World Are Examining The Source Code For Themselves. Some Have Already Removed The Restrictions And There is Another Version Already Available, With It's New Source Code, That Has The Restrictions Removed!+!! The Only Software You Should Really Trust is "Open Source Software". Note: Free Software is Not "Open Source Software" If You Don't Have Access To The "Source Code". I Have Paid For "Open Source Software", Where The Author Has Released The Sourcode to Be Examined. Eg. Some Linux Software.

asking BYCLOUD for face reveal day 1

China did open source it to slow down OpenAI. They have no such easy access to hardware yet. They need thier own and thier own suck so what they did they made OpenAI monopoly a past by open source. Now anyone can use deepseek to start thier own service whatever it will be on thier own hardware and China can in time catch up with hardware.

When the title is technical breakdown why waste the first 5 mins in lame introd

Every country fighting for who has better as Ai: hehehe, going as planned

Subscribed. Thanks for the dense technical coverage.

As someone doing safety testing, it would be a wet dream to be able to see exactly what data they use so that one could maybe even correlate behaviours with that. Sadly almost no one does it 🤧

The best thing about DeepSeek is not just the low cost and non reliance on CUDA, but the open source nature, They built on top of previous models like Llama and Qwen and gave back so people can build upon them as well. If someone distilled R1 or made a better version, they wouldn't complain that they are being copied but rather promote it as benefit of science.

I did a dissertation on Deep RL. To see RL algorithms like GRPO, PPO, A3C, Truncated Quantile Critics applied to LLMs for optimization is amazing to me.

What data is DeepSeek R1 AI trained on? How did the company acquire that data?

advertisement ends at 2:12

"We've achieved AGI. Wait! We've redefined AGI so we can keep getting money from Microsoft."

Thieves accuse others 'stole' from them.

you have many good and correct takes but on "stolen" data - i think there should be a lot of nuance and latitude given to Sacks' comments thus your rebuttal may be too shallow. you can still legally make API calls to the 01 model by paying fot he sbuscription and just manually or automatically max out your prompts against their EULA with prompt farmers and use that either for supervised training or orreinforcement training aggregating and even synthesizing the outputs and feeding it back into v3 to train R1 Everything prior to 01 like GPT4 or 3.5 turbo etc etc, whatever method they use to extract knowledge as part of their pool of MOEs, whichi s over 150+ we'll never know so you can't rule out the term "stolen" nor can you rule it in definitively.

USA 🇺🇸 is still so arrogant . Never underestimate 1.4 billion Chineses potential.

0:41 🤣🤣 Ah, NYC!

Great breakdown

Deepseek sucks. It's like a chatGPT 3. Newspapers told it was a revolution, it is not, try it for your own. It probably steals your data, it can't process any file format besides the text files, it's censored, it crashes 40% of the time, can't create images or videos, it's slower, gives inaccurate responses (more than chatGPT at least), can't recall old chats, lacks any type of team functionality, can't search on the internet without you pressing the actual button and it's actually slower than chatGPT and it's repetitive, so it gives you the same response over and over. And if this isn't enough, they copied the entire thing. You can argue they didn't copy the backend, but listen: - The frontend is copied, no one can deny it, and why violate copyright laws only for the frontend? The most important part is the backend, I would rather copy that part or both, but not only the frontend. - It gives the exact same responses chatGPT 3 did. - The filter for the censorship is clearly added later, and wasn't meant to be there at the beginning: in fact, if you ask about tinnamen square, it first gives you the response and then cancels it out. If they trained it in China, why apply the filter later? You could apply it before it gives you the response; when it receives the prompt; It could tell you nothing happened in tinnamen square. - And the most important one: If it is in fact trained in China, how does it know what happened in tinnamen square at all? Theyr internet has not any information regarding it, so it's clear it was trained in a western country, which doesn't make sense since electricity in china is like half the price, so they probably stole the training files from openAI when the best model was chatGPT 3, then they re-adapted it to the Chinese censorship, built the datacenter and now it's live (and sucks).

Why did DeepSeek and ChatGPT go to a party together? Because DeepSeek wanted to show ChatGPT how to have fun without costing an arm and a leg!😄

4:49 bro, 6.4M at 87% efficiency is not 5.78M, it would be cheaper to be less efficient?