DeepSeek, China's AI model: News & Discussion

[r3alist]

One must see the philosophy behind Deepseek differently. From what I gather its purpose is to allow smaller companies, individuals with lower end infrastructure to run a top end model. Unlike chatgbt4o/1o etc they dont need all the compute power to try and cover every topic and have it learn everything like the encyclopaedia, they can focus on the topics they want like medicine, finance etc.

Having it open source and with so many different models makes this accessible to so many people now, it breaks a monopoly the big tech firms have had as you dont need to spend $millions in setting up decent AI models for your own purposes. Ofcourse the AI models being backed by the billion $$ firms are no joke either, but with what deepseek has done they can probably make their models leaner and more efficient with all that compute behind them.

Well it's proven that the focus on hardware was ignoring that you can be more efficient with the software and get as good as the industry benchmark of chatgpt without big investment, that will cause disruption but in fact also a focus on optimising more aggressively, I guess everyone was getting a piece of the hype, along the supply chain


Ok.... But ultimately we are all around the level of chat gpt , so it's an actual game changer if you really can make use of chatgpt right now, they made it open source because it also undermines the investment and supply chain industry in the USA

Is there a real level up in capability? Ultimately won't compute beat software refinement? Who knows what djrei this goes

 

[Hamartia Antidote]

You think chatgpt doesn't restrict certain topics

 

[F-22Raptor]

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

100%, this is what ultimately matters

 

[F-22Raptor]

Well it's proven that the focus on hardware was ignoring that you can be more efficient with the software and get as good as the industry benchmark of chatgpt without big investment, that will cause disruption but in fact also a focus on optimising more aggressively, I guess everyone was getting a piece of the hype, along the supply chain


Ok.... But ultimately we are all around the level of chat gpt , so it's an actual game changer if you really can make use of chatgpt right now, they made it open source because it also undermines the investment and supply chain industry in the USA

Is there a real level up in capability? Ultimately won't compute beat software refinement? Who knows what djrei this goes

See above

Frankensteining software will only get you so far. Eventually you hit a wall without the compute

An the US has a clear moat in compute.

 

[SaadH]

This whole bunch of cr*p about US being attractive to "talents" was always nonsense as once you get above a certain population threshold like the billion+ with China, then it is all about your culture and system in order to give the opportunity to those rare geniuses to thrive.

China's system(education, cultural and governance) is on another level of efficiency compared to the US.

It is now in a race with itself for world technological dominance.

Chinese system, economy and it's STEM culture is devoid of financization, whereby everything is looked at through the prism of maximizing profit and delivering value only to the zionists on wall street. There is no real striving to make scientific breakthrough or worst yet offering facility and benefit to the public at large.

Chinese deliberately made the effort to block this malaise that plagues the US economic system where talentless hacks with no scientific and/or engineering background get to run trillion dollar companies and horde hundreds of billions of dollars just to satisfy their selfish narrow egos.

 

[SaadH]

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

100%, this is what ultimately matters

Powerplant/engine is ultimately what matters in a fighter jet. Strapping 20 jet engines to a fighter jet is not going to solve all the problems.

 

[F-22Raptor]

Powerplant/engine is ultimately what matters in a fighter jet. Strapping 20 jet engines to a fighter jet is not going to solve all the problems.

Compute is the ultimate enabler of AI.
Frankensteining software will only take you so far.

Aircraft are a poor comparison as they are ultimately constrained by the laws of physics. You can only make an aircraft so large to achieve your objective. Aerospace engineers are left with no choice but to improve the efficiency of its engines.

Compute OTOH is in its infancy, that’s why the American private sector is investing hundreds of billions in data centers. To achieve ASI, you need massive compute. Efficiency alone won’t get you there, hence compute being the enabler.

 

[F-22Raptor]

Compute is the ultimate enabler of AI.
Frankensteining software will only take you so far.

Aircraft are a poor comparison as they are ultimately constrained by the laws of physics. You can only make an aircraft so large to achieve your objective. Aerospace engineers are left with no choice but to improve the efficiency of its engines.

Compute OTOH is in its infancy, that’s why the American private sector is investing hundreds of billions in data centers. To achieve ASI, you need massive compute. Efficiency alone won’t get you there, hence compute being the enabler.

R1 is a good example. Deepseek is compute/advanced chip constrained, which is why they had to focus on efficiency. Efficiency got them to near o1 level performance, but not leading edge performance of o3 etc.

We’ll see how well China competes when US AI companies begin using millions of Blackwell/Rubin GPU clusters in its data centers.

 

[F-22Raptor]

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

Deepseeks efficiency is being massively oversold

 

[Developereo]

LOL, you are so desperately panicking you contradict yourself in the space of just two posts:

In the first post, you claim American AI will win because it has access to massive hardware compute power which is what really counts.

To achieve ASI, you need massive compute. Efficiency alone won’t get you there, hence compute being the enabler.

And then you trash DeepSeek because, you say, it relies too much on hardware compute power.

Deepseeks efficiency is being massively oversold

 

[j_hungary]

LOL, you are so desperately panicking you contradict yourself in the space of just two posts:

In the first post, you claim American AI will win because it has access to massive hardware compute power which is what really counts.



And then you trash DeepSeek because, you say, it relies too much on hardware compute power.

DeepSeek was popular because of their efficiency, it show that AI don't need multiple hundred million dollars (think DeepSeek was made with 6 mil?) with multiple server to actually push the program forward, it impacted Nvidia because that's what they do now, almost all Nvidia chips are for big brand AI Company that pushes the envelope. Which DeepSeek just show you don't need them to have a proper AI network. And also, DeepSeek is opensource, so that's basically SME dream because you don't need to pay hefty license fee to use them.

On the other hand, DeepSeek is more toward the functioning side of AI (or strictly application AI) which is good for like your phone (Process Photo, managing contact and so on) and while you will still need big server for Machine Learning model AI

DeepSeek-V4 Pro vs GPT-5.5 - Detailed Performance & Feature Comparison

Discover how DeepSeek's DeepSeek-V4 Pro and OpenAI's GPT-5.5 stack up in performance, features, and applications. Read our detailed comparison to find out which AI model best suits your needs.

docsbot.ai

 

[RealNapster]

Deepseek is said to have same performance as Chatgpt o1. While OpenAI charge you for using o1, Deepseek is free to use and business can use it's open source models to build over it.

To a consumer, that's all what matters. It doesn't matter if Deepseek will not answer about tianmen square incident. Why does it matter ? Can it provide accurate and authentic responses on STEM is all it matters to researchers. Regarding the data theft by Deepseek, tell me with a straight face that CIA , META, Microsoft, and OpenAI is any different in this regard.

 

[j_hungary]

Deepseek is said to have same performance as Chatgpt o1. While OpenAI charge you for using o1, Deepseek is free to use and business can use it's open source models to build over it.

Yeah, this is why DeepSeek is interesting.

It's like someone created a port for you to play PS3 Games but in a very small system and is free, this is what so appealing to SME (Small and Medium Enterprises) it's good for gamer in general if you just want to play games because you pay nothing to have access to it. You don't need top of the crop platform if you are doing some enterprising application (like analyse customer behavior) and you don't need to pay huge amount of money to incorporate AI in your work. Because it's FREE........

 

[Developereo]

Regarding the data theft by Deepseek, tell me with a straight face that CIA , META, Microsoft, and OpenAI is any different in this regard.

Even on Australian websites people are laughing at the scare campaign about DeepSeek stealing data. People are saying "What do I care if my data is stolen by CCP or Google? It's stolen either way. Is CCP going to spam me with ads?"

 

[Faithfulguy]

I think everyone needs to chill, I don't believe the Chinese are being hugely triumphalist even, it's more the west reminding itself it can't lose to china so creating it's own hysteria, you think they don't have more capability in the West lol

The big tech mag 7 needed a correction anyway

This China Vs USA will go on for years, this is the first sugar rush


And rest assured we have Tata consultants ready to lend their expertise on qa projects

hahaha. QA jobs would be the first to go as AI is the most efficient at creating test code vs development code.