DeepSeek, China's AI model: News & Discussion

[大汉奸柳传志]

Maybe you should act like a contributing forum member instead of the old guy in the stands with the shaking cane that just heckles people on the playing field.

Have you ever thought about that?

oh shut your talking hole Peter Clueless from MA..

I am contributing by setting you straight, consider yourself soft banned from participating in AI related discussions until India makes its own top AI model.

 

[Hamartia Antidote]

oh shut your talking hole Peter Clueless from MA..

haha...getting easily triggered...like an Indian.

Back on topic please....let's keep it quiet in the peanut gallery.

 

[大汉奸柳传志]

UK just wants to be the middleman of China-US trade during Trump years, taking the role from Vietnam, Malaysia and to some extent Mexico.

I bet Trump will never raise tariffs on UK

 

[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

 

[ety]

I bet Trump will never raise tariffs on UK

They are best buddies along with Israel, otherwise he is completely mad. Tariffs on EU isn't too likely too.

 

[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

 

[Hamartia Antidote]

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

I posted earlier that the CEO mentioned in an interview having access to 50,000 nvidia H100's and at $25,000+ each that ain't cheap.

However if their servers are running the inference on cheap Huawei chips then that could be low cost for users. They could just train it with the H100's.

 

[F-22Raptor]

I posted earlier that the CEO mentioned in an interview having access to 50,000 nvidia H100's and at $25,000+ each that ain't cheap.

Deepseek has clearly lied about the total costs and stole OpenAI IP on top of it.

Deepseek are frauds

 

[GiantPanda]

If only China can translate its AI success into the biotechnology / Medical field, offering / inventing / discovering new treatments and medications for untreatable diseases, that would be revolutionary.

It is already happening: cures for diabetes and alzheimer.




This is only the beginning. China is doing this when it is still a developing nation. You are talking about a teenager in national development terms.

Once China matures, it will dominate science and research like no other countries before.

 

[大汉奸柳传志]

one of the many unique things about DS that people fell in love with

Deepseek has clearly lied about the total costs and stole OpenAI IP on top of it.

Deepseek are frauds

All I see from your post is

I posted earlier that the CEO mentioned in an interview having access to 50,000 nvidia H100's and at $25,000+ each that ain't cheap.

just leave the thread.

 

[tower9]

Resist what? Britain is a consumer not a producer of tech.

 

[Beijingwalker]

DeepSeek Technology Is An Ace Up China’s Sleeve For Drone Warfare​

David Hambling
Senior Contributor
Jan 29, 2025,09:18am EST

Chinese PLA soldiers practice with FPV drones

CLASH REPORT VIA TWITTER
China’s release of the DeepSeek chatbot has sent shockwaves through the U.S. stock market and beyond. The makers claim that the underlying DeepSeek-V3 model was trained for around $6m, a tiny fraction of the cost of rival U.S. systems, and with a tiny fraction of the computing power.

This happened just after the U.S. had announced massive spending on AI infrastructure to build the sort of computing power which DeepSeek’s appears to make questionable. The launch has been described as ‘Sputnik moment’ echoing events in 1957 when the Soviet Union shocked the world by taking a lead in the space race when they put the first satellite into orbit.

Sputnik had obvious military implications. Americans were nervously aware of the Russian hardware now flying over their heads every ninety minutes. A Chinese lead in Large Language Models may look like a purely economic edge. But a new generation of AI-enabled drones could translate it into a military advantage.

Putting An AI Crown On Drone Supremacy

DJI Sky City, the global headquarters of the world's leading consumer drone marker DJI, in Shenzhen.

VCG VIA GETTY IMAGES
China is the world’s largest producer of small drones. A single company, DJI of Shenzhen, commands an estimated 80% of the global consumer drone market. Although the company officially deplore military use of their product, DJI drones have become ubiquitous tools used by both sides in the war in Ukraine.

Low-cost FPV drones, armed racing quadcopters, have become perhaps the most important single weapon in the ground war. These are assembled locally but many or most of the parts come from China.

At present, almost all drones are flown by remote control. Skilled FPV pilots can score remarkable successes: 25-year-old Tymofiy Orel was awarded the title ‘Hero of Ukraine’ for destroying 42 Russian tanks and 44 armored personnel carriers among other vehicles.

In the past year or so small drones have been equipped with increasing levels of AI. In particular, U.S. makers Auterion are supplying Ukraine with Skynode-S boards, computers the size of a credit card optimized for machine-learning apps, for example locking on to a target and automatically flying the best route. Ultimately this could turn every drone operator into a Tymofiy Orel.

In an address to RUSI in October, former Ukrainian Defense Chief Valerii Zaluzhnyi noted how quickly the technology has evolved.

“When robots entered the battlefield [in 2022], there was laughter from the Western press and local generals,” he said. “However, by 2024, technology, particularly Artificial Intelligence, began to play a significant role in warfare.”

The next year is expected to see increasing numbers of AI-enabled drones on the battlefield in Ukraine. There are a number of challenges, including packing enough computing power on a small drone to run AI applications at speed, and training machine-learning systems to run reliably and efficiently.

This is just the sort of technology that China appears to have gained with DeepSeek.

From Virtual World To Battlefield​

DeepSeek is based on a Large Language Model (LLMs), a specific type of AI trained on large amounts of text and excels in carrying out conversations in natural language. This might seem a world away from the challenge of flying a drone, but researchers have shown how LLMs can enhance drone operations. Much of this research comes from China.

A 2024 paper provides a good overview of the field -- “Large Language Models for UAVs [Uncrewed Aerial Vehicles]: Current State and Pathways to the Future,” was written by Shanghai-based researchers who describe how LLM integration with drones can improve autonomous data processing and rapid decision-making.

While they carefully avoid specific applications, the authors note that “LLMs are exceptional at identifying specific objects, individuals, vehicles, or activities in video streams or images, providing detailed insights crucial for military and civilian surveillance operations.”

This translates into an ability to carry out complex commands like “see if there are any vehicles on the track by the river” or “"fly at low altitude to the enemy position and approach from the South. “

The LLM allows the drone to accept orders, then report what it sees in natural language: “no moving vehicles on the track, one burned-out truck as previously reported” or “there was no fire from the enemy positions, but heat sources show they are occupied.”

As well as communicating with humans, LLMs enable the drones to talk to each other in natural language. A group of drones can share findings and co-ordinate their actions based on their mission, and distributing tasks without human intervention. A reconnaissance drone might assign attack drones to specific targets, then send follow-ups on the ones which are not destroyed.

Another Chinese paper from 2024 takes this further – Manned and Unmanned Aerial Vehicles Cooperative Combat Framework Based on Large Language Models describes work using LLMs to make “ fully autonomous aerial unmanned combat aircraft “ a reality. They envision a single human fighter pilot with a collection of robotic wingmen who react at machine speed and communicate in human language. Again, they leverage the planning and problem-solving abilities of the LLM to make it an effective pilot, able to respond rapidly to unexpected situations. (It would be weird but not surprising if the drones imitated human practice and gave themselves callsigns like ‘Maverick’ and ‘Iceman’).

Leveraging The DeepSeek Advantage​

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

DeepSeek is claimed to have specific advantages over other LLMs, in particular that it is resource-efficient, meaning it needs less computing power than other models and suitable for edge applications like drones rather than data centers. In addition, the developers claim to have achieved results rapidly with less training data than comparable models.

We do not know how capable DeepSeek really is. But a powerful, low-cost system that can be deployed on drones would be a useful asset for China at this point.

China has the world’s biggest drone manufacturing base. They have a wealth of information about using these drones in combat with the tactics and techniques worked out by Ukraine and Russia in a war which has cost them nothing. And they have recently being developing their own drone combat units equipped with what look like low-cost commercial drones.

Add LLM capability to the mix and China seems to hold all the cards for building an unbeatable drone force. Warfare is becoming increasingly a matter of drone vs drone. In the long run, the software may turn out to be the deciding factor as victory will go to the side with the smartest drones.

 

[UKBengali]

At least ChatGPT gives an answer about Gaza...while DeepSeek played the "safe" it "wasn't aware of anything" game

Meanwhile ChatGPT...

Are you aware that Deepseek knowledge base is only up to October 2023?

Of course it cannot answer that question!

 

[Hamartia Antidote]

Are you aware that Deepseek knowledge base is only up to October 2023?

For https://chat.deepseek.com

Of course it cannot answer that question!

 

[SaadH]

DeepSeek R1 has essentially destroyed the American monopoly over AI before it even began. The reason being cheap asf inference and training costs, and it being open source. Now most people have been clinging to the Tianeman Square and Xi Jin Ping questions but miss out on a major point. This censorship applies only to the model interface hosted within China so of course whatever the CCP approves of, the model will be allowed to talk about. Similarly, OpenAI ChatGPT does not want to talk about the Rothschild family at all. Now the point where DeepSeek takes the cake is it being open source. A person with the know how could adjust the weights and remove any and all censorship from the model. It is available to everyone to use for free and for any application. You can host it offline on your own hardware so that no data goes to China and neither does any data get censored. Americans are trying to downplay this HUGE blow to their billions of dollars invested into AI because a minor company has come and destroyed their flagship reasoning model for a fraction of the cost.

And for people still skeptical about it, Hugging Face is currently cloning (re training from scratch with help from DeepSeek itself) the model in its entirety so that it is free of any censorship, and hosted in America. Perplexity already has done that but with minor censorship. OpenAI essentially has to produce a miracle such that a better model than R1 is made and is made open source just like it, else why would any person in their right mind pay for O1 or O3 when they can just host it on personal hardware for a fraction of of the cost.

Also, this model introduced a new type of compute, and hence is groundbreaking technology. It is using Reinforcement
Learning unlike O1 that used supervised learning. Reinforcement will always be superior to supervised learning so until OpenAI works on a RL based model, R1 is king.

With $6K investment in hardware you can even run it locally off your own PC.