DeepSeek, China's AI model: News & Discussion

[vasanthm]

That's because they were not founded by Indians. Being a CEO is not the same as being a founder of a company

Sabeer Bhatia (Hotmail), Vinod Khosla (Sun Microsystems), Dheeraj Pandey (Nutanix), Aneel Bhusri (Workday)—all Indian founders, but no one calls those “Indian” companies. It’s not just about the founder’s nationality; it’s about the ecosystem, resources, and opportunities a country provides that shape a company.

You Chinese have a strange way of looking at things. Sure, we have our own idiots claiming everything as “Indian,” but you guys take the delusion to another level!

 

[F-22Raptor]

Talking none sense
View attachment 97568

o1s AIME performance is actually 83.3, so yes R1 is an inferior product to o1.

But again o1 is an old OpenAI LLM, they were done with o3 a couple months ago and it far exceeds R1 in performance.

o1 is old news, OpenAI is already training o4 and beyond

 

[Hamartia Antidote]

It's founded by a Chinese guy so yes Nvidia is a Chinese company u have a problem with that? Same for twitch , Yahoo, YouTube, doordash, lam research and many more lol

Let's ask DeepSeek some questions...since it wouldn't have any Western bias....

To view this content we will need your consent to set third party cookies.
For more detailed information, see our cookies page.

Accept third party cookies

 

[liuzhengdong]

Deepseek is a ChatGPT reskin, and they’re lying about the costs. And nearly every major American AI CEO has called them out on it, including Elon.

The capitalists on Wall Street are not as stupid as you. Stocks can't deceive people.

 

[FHN]

Let's ask DeepSeek some questions...since it wouldn't have any Western bias....

View attachment 97573
View attachment 97569

To view this content we will need your consent to set third party cookies.
For more detailed information, see our cookies page.

Accept third party cookies

View attachment 97570

View attachment 97571

View attachment 97572

Ok he was the first employee. What are u trying to tell me? Jensen founded Nvidia in 1993 and remains the CEO ever since. Twitch used to be called justin.tv name kinda cringe tho

 

[RealNapster]

NVIDIA is not a Chinese company. It was founded in California. Jensen got all his education in the US. And the fact the US controls the most advanced GPUs and its exports just confirms that.

Open source or not, R1 is an inferior model to o1. And o3s benchmarks far exceed that of o1

Now cope

Your stock exchange lost $1 trillion to an inferior chatgpt replica ?

 

[FHN]

o1s AIME performance is actually 83.3, so yes R1 is an inferior product to o1.

But again o1 is an old OpenAI LLM, they were done with o3 a couple months ago and it far exceeds R1 in performance.

o1 is old news, OpenAI is already training o4 and beyond

Cope more boy facts don't lie
DeepSeek R1 outperforms OpenAI o1 in key benchmarks:
- **AIME 2024 (math)**: 79.8% vs. 79.2%
- **MATH-500 (advanced math)**: 97.3% vs. 96.4%
- **SWE-Bench (coding)**: 49.2% vs. 48.9%
But o1 narrowly wins in **Codeforces (96.6 vs. 96.3 percentile)** and **MMLU (general knowledge)**.
DeepSeek R1:
- Input: $0.14/million tokens
- Output: $2.19/million tokens
- OpenAI o1:
- Input: $7.5/million tokens
- Output: $60/million tokens
For startups and researchers, R1’s pricing is a game-changer. It takes 2 months for deepseek to develop r1 as a side project.

 

[liuzhengdong]

o1s AIME performance is actually 83.3, so yes R1 is an inferior product to o1.

But again o1 is an old OpenAI LLM, they were done with o3 a couple months ago and it far exceeds R1 in performance.

o1 is old news, OpenAI is already training o4 and beyond

You are disparaging DeepSeek, but you should explain why Nvidia's stock price plummeted. Are those Wall Street elites dumber than you?

 

[AViet]

Always believe this day will come.

 

[F-22Raptor]

The capitalists on Wall Street are not as stupid as you. Stocks can't deceive people.

Thats just a panic sell. Stocks fluctuate irrationally. That was an irrational panic sell. With millions of GPU clusters coming in the years ahead for AI compute, NVIDIA is fine

Not even Elon take Deepseeks claim seriously:

To view this content we will need your consent to set third party cookies.
For more detailed information, see our cookies page.

Accept third party cookies

 

[F-22Raptor]

Cope more boy facts don't lie
DeepSeek R1 outperforms OpenAI o1 in key benchmarks:
- **AIME 2024 (math)**: 79.8% vs. 79.2%
- **MATH-500 (advanced math)**: 97.3% vs. 96.4%
- **SWE-Bench (coding)**: 49.2% vs. 48.9%
But o1 narrowly wins in **Codeforces (96.6 vs. 96.3 percentile)** and **MMLU (general knowledge)**.
DeepSeek R1:
- Input: $0.14/million tokens
- Output: $2.19/million tokens
- OpenAI o1:
- Input: $7.5/million tokens
- Output: $60/million tokens
For startups and researchers, R1’s pricing is a game-changer. It takes 2 months for deepseek to develop r1 as a side project.

R1 is an inferior model to o1 on parameters and vastly inferior to o3 capability.

And Deepseek is training off of American GPUs.

Cope

 

[UKBengali]

o1s AIME performance is actually 83.3, so yes R1 is an inferior product to o1.

But again o1 is an old OpenAI LLM, they were done with o3 a couple months ago and it far exceeds R1 in performance.

o1 is old news, OpenAI is already training o4 and beyond

You are missing the point.

Deepseek is far more efficient as it matched OpenAI for what it claims is 5% of the cost.

They are of course developing even more advanced models(most probably now being trained on Chinese Huawei Ascend 910C GPU) that are in the works.

Deepseek simply is far more efficient in "training" these AI models than either Google or OpenAI.

 

[FHN]

R1 is an inferior model to o1 on parameters and vastly inferior to o3 capability.

And Deepseek is training off of American GPUs.

Cope

The only one coping here is you. I ain't got all the time to respond to you. The charts I show you speak for themselves whether you believe it or not is up to you. GN my Vietnamese boi

 

[F-22Raptor]

You are missing the point.

Deepseek is far more efficient as it matched OpenAI for what it claims is 5% of the cost.

They are of course developing even more advanced models(most probably now being trained on Chinese Huawei Ascend 910C GPU) that are in the works.

Deepseek simply is far more efficient in "training" these AI models than either Google or OpenAI.

Again “claims”, a claim in which not even Elon believes.

The future of AI comes down to compute and millions of GPU clusters. Thats why the US is at the frontier in both hardware and models and in the best position. Deepseeks founder even recognizes their biggest problem: Advanced chips, and the Ascend series is far inferior to NVIDIA products

To view this content we will need your consent to set third party cookies.
For more detailed information, see our cookies page.

Accept third party cookies

To view this content we will need your consent to set third party cookies.
For more detailed information, see our cookies page.

Accept third party cookies

R1 is an inferior product to o1 and o3 vastly exceeds it in capability.

Funding is not the issue for American leading edge AI companies. Compute was the issue and that’s being solved with hundreds of billions of dollars invested into data centers.

American AI advances will vastly accelerate from here.

Deepseeks problem is access to advanced chips/compute. It will never exceed American frontier edge AI capabilities without it.

 

[UKBengali]

Again “claims”, a claim in which not even Elon believes.

The future of AI comes down to compute and millions of GPU clusters. Thats why the US is at the frontier in both hardware and models and in the best position. Deepseeks founder even recognizes their biggest problem: Advanced chips, and the Ascend series is far inferior to NVIDIA products

To view this content we will need your consent to set third party cookies.
For more detailed information, see our cookies page.

Accept third party cookies

To view this content we will need your consent to set third party cookies.
For more detailed information, see our cookies page.

Accept third party cookies

R1 is an inferior product to o1 and o3 vastly exceeds it in capability.

Funding is not the issue for American leading edge AI companies. Compute was the issue and that’s being solved with hundreds of billions of dollars invested into data centers.

American AI advances will vastly accelerate from here.

Deepseeks problem is access to advanced chips/compute. It will never exceed American frontier edge AI capabilities without it.

Elon Musk is not the *truth*.

Again you are missing the point completely.