Видео

DeepSeek iOS app sends data unencrypted to ByteDance-controlled servers but if you run it locally then no worry

some see it as a wake-up call exposing a free market undermined by government favoritism toward entrenched corporate oligarchs, while others question its methods and sustainability.

Yh they will not public them ( they don't have such history) But eu and usa will do. I didn't think we survived in tech products for now without eu and usa as most of their own house. Mini unnecessary apps ,web usage

It's a provocative claim, but definitive evidence that DeepSeek repurposed OpenAI’s SFT outputs using RLHF remains unverified.

It sure looks like America's heavy-handed approach in AI has backfired, proving that even the biggest players can get spanked when innovation turns the tables.

thanks for suggestion im just exp with diff open source .

simply skip it

leaner AI innovation can truly benefit society rather than just enriching the ultra-rich.

100%

ah ha movements 🤣🤣

not sure as most of them are still paid research work . but its new methods

While ownership and regional adoption play a role, what excites me most is how this race pushes AI to benefit users everywhere.

yh they maybe updated it considering the dominance of R1

Absolutely! Open-source research drives innovation and benefits everyone by fostering competition. DeepSeek’s paper is a great example of how sharing knowledge can push the entire AI field forward. More competition means better, more affordable technology for everyone!

...its an ancient Chinese principle of engagement... Sun Tzu

Well greedy corps don't want competitions so they can charge you a fortune.

research should be open source as yunn lecun emerge it more .

By leveraging a new approach called PTX, DeepSeek can bypass CUDA to some extent. Instead of using CUDA, it utilizes NVIDIA's PTX (Parallel Thread Execution) language for implementation. PTX operates at a level closer to assembly language, allowing for fine-grained optimizations such as register allocation and thread/warp-level adjustments. However, this programming approach is extremely complex and has low maintainability, which is why the industry typically prefers higher-level languages like CUDA. In other words, they have pushed optimization to the extreme. One commenter remarked, "Those who abandoned CUDA for PTX because they found it slow were probably former quantitative traders." It’s important to clarify that PTX is still a part of NVIDIA’s GPU architecture, serving as an intermediate representation within the CUDA programming model. Its role is to bridge high-level CUDA code with the GPU’s low-level hardware instructions. Therefore, while DeepSeek has optimized at the PTX level, this does not mean it has completely detached from the CUDA ecosystem. Rather, it demonstrates DeepSeek’s ability to deeply optimize GPU performance.

@@CodewithRizI suppose CUDA too can be similarly extremely optimized bringing about unimaginable scalability?

i partialy agree with last line .

I understand your concerns about data usage and transparency. DeepSeek has published their methodology openly for independent review, which is meant to foster healthy competition and innovation, but I agree that ongoing investigations will hopefully clarify any issues. Ultimately, whether one supports or refrains from using these tools is a personal choice, and I appreciate the discussion on how data and technology should be responsibly shared.

when a pressing need or problem arises, it drives people to seek solutions through scientific exploration and innovation, essentially implying that the need itself acts as the catalyst for scientific advancement

when a model delivers nearly the same performance but at a fraction of the cost, that cost-efficiency can be a true game changer. Even if DeepSeek’s investment were scaled up tenfold to, say, $56 million, it's still minuscule compared to the astronomical amounts (on the order of tens or hundreds of billions) that giants like OpenAI and Microsoft spend, which can make the technology much more accessible and disruptive.

efficient AI model like DeepSeek’s challenges the idea of full spectrum dominance-the notion that a single, dominant power or technology controls every aspect of the field. By making advanced AI accessible without the huge investments required by traditional models, DeepSeek disrupts that monopoly, potentially undermining the control that major players have over the entire AI landscape. This could lead to a more competitive, decentralized environment rather than one dominated by a few giants.

1. OpenAI stole from tons of sources that had terms and conditions against it 2. About 80%+ of openai's code is open source code. The T in GPT comes from Transformers which is what everything is based on which was created open source by Google, using tons of tools open sourced by meta, with methods from labs all over the world. 3. The theory is that maybe R1 used destillation of o1, stealing code is out of the question, but even that it's not proven and even that it's commonly done by other open source models like Meat's 4. Deepseek evolved and shared tons of new research, they implemented several open source research, including from Google and Mistral (french LLM lab) evolved, and shared it contributing to the world's knowledge.

@MrEnriqueag it's going to be spyware just like everything else China makes. They don't have best interests for anyone, and they have proven it time and time again. They are the worst human rights abusers on the planet. This DeepSeak is going to blow up in everyone's face just like everything else China pretends to do better then everyone else.

what do you think about balaji claim ,how they use our data to rain gpt ? scinece would be open source as its benifits for all of us (will you spent 200$ for same quility they provide )

I know China is one of the most egregious violators of human rights. That every time they say they've built something cheaper and better than us it actually turns out to be shit or a scam. 🤷

Maybe, and maybe this is all a huge shit show of smoke and mirrors for market manipulation. Who knows?

Well there are several factors here shareholders know more than we know thats why NVIDIA lost 600 billion and well we don't know what tech can actually do we are just using various techniques like they tried in terms of developing AI model one worked out in terms of cost with just 10% performance tradeoff

R1 is trained through reinforcement learning to understand the patterns of other ai models. Instead of being trained on millions of documents made by humans it's trained on the responses of models like o1. Like a student learning from a master instead of on their own. That's why it's cheaper to train. But theres a big problem with that: it can't be much better than the models it's trained on, and if people stop developing and training models on real world data, R1 dies too. Edit: Also, R1 is much more susceptible to hallucinations than other established models... Hallucinations are the most important problem to solve.

regardless of how you feel, its better than chat GPT whether they copied or not! which we have no proof of also its free to use, so be grateful and dont be overcome by prejudice

@@polblanes the fact is that R1 is better than O1 ChatGPT is trained on stolen human inputs. The transformer model was published by researchers at Google. You can't replicate others and call it innovation and simultaneously call others out for doing the exact same thing. DeepSeek published their methods and open sourced their models, and OpenAI is currently replicating it. OpenAI claimed to be "open" and "non-profit" but is actually "closed" and "commercial". and the reverse is actually true, if developers stop spoon feeding gpt, o1 is dead

ah man , look its ah ha movements 🤣🤣🤣

Your skepticism about DeepSeek's authenticity is understandable. Recent reports indicate that OpenAI suspects DeepSeek of using ChatGPT data to develop its AI model, raising concerns over AI ethics. also looked how gpt was trained (mean where they gets the data)

Ask Western chatbots about Topics they don’t like. It’s the exact Same thing. It’s the most stupid thing to say stuff like that when It’s just two sides of the Same Coin

@@TheBubble010 exactly

Ask OpenAi how to overthrow the US government and have all states to be independent and see what you get

Innovation is about making the best use of available resources, and DeepSeek has done just that. I appreciate their approach-copy, optimize, and make it more affordable!

start from small . its depends on whats data you have

lora is fine tune methods which helps to optimize your training time only nothing else .

you can use deepseek coder for coding language fine tunning they have spacifics models for these task . speed is still depend on hardwares .

@@CodewithRiz thanks bro

You can try others as it's my own experiment. With text you must have to know facial expressions details along prompt expertise