Artificial Intelligence thread

[Wrought]

SOEs are directed to contribute to open-source development and produce more 0-1 innovations, as well as accelerate commercialized AI applications. Obviously there's lots of attention on big tech companies, but most of the economy is in less flashy stuff. Good to see government encouraging broad adoption.

Please, Log in or Register to view URLs content!

 

[Overbom]

DeepSeek now used in ports where it significantly speeds productivity

Please, Log in or Register to view URLs content!

The Ningbo-Zhoushan Port – the world’s third-largest container port by throughput in 2024 according to shipping news service Lloyd’s List – has begun integrating the DeepSeek model into its intelligent gate and customer service system.

The new gate system can recognise over 80 per cent of containers’ unique seal numbers, significantly reducing the time needed for manual verification. Container inspections that previously required four staff members can now be handled by a single person, according to the port’s owner.

The AI model also helps the gate system reduce its fault alert response time to within half an hour. The DeepSeek-backed gate has been installed at multiple terminals, processing 20,000 20-foot equivalent units a day.

Zhoushan Port is also trying to integrate DeepSeek’s model into its 24-hour customer service system, which is already supported by

Please, Log in or Register to view URLs content!

, a general-purpose large language model developed by Alibaba – the tech giant and owner of the South China Morning Post.

Customer service applications of the model are currently testing and scheduled to launch in the first half of the year. The port said it has plans for other, expansive applications, including cargo handling, gantry cranes and security systems.

Guangzhou Port, in the capital of southern China’s Guangdong province, has integrated the DeepSeek model into its equipment management and billing systems to enable faster responses to malfunctions and boost data input accuracy.

And Hubei Port Group is constructing a platform supported by AI models, including DeepSeek, to oversee logistics operations across the whole of Hubei province. It is also exploring AI applications for its new multimodal terminals at Yangluo Port in Wuhan

Please, Log in or Register to view URLs content!

.

 

[Sinofan]

Please, Log in or Register to view URLs content!

Hi. Can some experts help to explain the significance of
1) the slow speed for Deepseek R1 relative to other models?
2) the higher price of 500% more for Deepseek R1 when compared to Gemini 2.0?

How will these factors matter for end-users and determine the eventual winning model?
Thank you very much.

 

[Biscuits]

That's exactly the question I've been asking. I don't know how Deepseek will benefit money wise from all this. I'm not sure they started it to grow and make money. If that's the case, I'm not sure what will be next for them, if they will continue growing their business in AI or other bigger Chinese AI companies will take over

How do you think he was going to make money? By sitting on the patent and collecting royalties?

Deepseek founder has been very transparent about his goal, to develop AGI. If he secures majority funding/mandate from the government to do that, he becomes the richest/most assets holding man in the world.

The point behind R1 is to raise public interest and prove to the powers above that he should be the chosen contractor to make AGI. It's obvious how Deepseek would vastly profit from that, not just in raw money but also data access. They could stand to gain multi trillion $ value depending on what mood the leadership is in.

 

[tphuang]

https://twitter.com/i/web/status/1893264072700813654

ByteDance Doubao has been overtaken by Tencent Yuanbao in popularity.

 

[tphuang]

Please, Log in or Register to view URLs content!

Oppo continuing with its develop of 阶跃星辰 Step multi modal model for different end points like smart cars, robots, mobile devices & IoT

 

[Hyper]

https://twitter.com/i/web/status/1893264072700813654

ByteDance Doubao has been overtaken by Tencent Yuanbao in popularity.

We might see the six tigers slide into irrelevance. Just like the previous startups before them. Baidu and others will now face great difficulties as this is another market failure for them. Alibaba will get that sweet Apple money.

 

[siegecrossbow]

DeepSeek now used in ports where it significantly speeds productivity

Please, Log in or Register to view URLs content!

This is the type of AI revolution that I’ve been talking about. Chatbot buy themselves do not provide anything revolutionary other than maybe making white collar work redundant. Only have you mated it to physical hardware or infrastructure can you perform truly meaningful feats.

 

[Overbom]

Please, Log in or Register to view URLs content!

Hi. Can some experts help to explain the significance of
1) the slow speed for Deepseek R1 relative to other models?
2) the higher price of 500% more for Deepseek R1 when compared to Gemini 2.0?

How will these factors matter for end-users and determine the eventual winning model?
Thank you very much.

Dunno mate, it looks wrong to me.

Slow speed is due to DeepSeek servers getting hammered, not due to the actual model being slow.
Check OpenRouter. Faster R1 provider is Fireworks with 72 tokens per second not 25 tps. However they are more expensive




Here is the actual details for Gemini 2.0. And notice the extremely limited output limit of only 8k tokens.



Nevertheless, I have long said that Google is one of the best if not the best positioned company in AI. Their Gemini models are extremely cheap and fast while their hardware is great for inference.

DeepSeek just needs to scale up, and if not, what's the harm, R1 is open source so other companies can scale it up if DeepSeek can't or won't

 

[tokenanalyst]

Tsinghua Unigroup releases Lingxi Cube large model all-in-one machine (DeepSeek version)​

Unisplendour Corporation stated on an interactive platform that H3C, a subsidiary of Unisplendour Corporation, recently officially released the Lingxi Cube large model all-in-one machine (DeepSeek version). The product includes two series, the Pure Enjoyment Edition and the Enablement Edition, with a total of 12 products in six models, covering DeepSeek large models ranging from 14B to 671B in size.

H3C provides a "hardware + model + full stack + service" service system, including early consultation (based on government and enterprise service experience, customized scenario-based implementation plans for enterprises), mid-term deployment (supporting elastic expansion and contraction and API encapsulation, dynamic adjustment of computing resources on demand, and improving utilization), post-operation and maintenance (through AI capabilities, improving automated operation and maintenance monitoring methods and efficiency), and value-added services (providing subscription services such as computing power optimization and model parameter adjustment to unleash the performance potential of AI) full life cycle services to provide users with localized deployment and services of the UniCube DeepSeek large model all-in-one machine.

H3C said that the DeepSeek all-in-one solution released this time not only meets the stringent requirements of enterprise-level scenarios, but also once again verifies the excellent support capabilities of H3C's intelligent computing infrastructure for large models, fully activating the intelligent potential of large models. In the future, H3C Group will continue to uphold the concept of "intensive cultivation and pragmatism, and empowering wisdom for the times", continuously accelerate the evolution of product technology, improve the ecological cooperation system, empower high-quality large models including DeepSeek, and accelerate the digital transformation of hundreds of industries.

Please, Log in or Register to view URLs content!