DeepSeek: All the News About the Startup That's Changing the AI World

DeepSeek, a Chinese startup, is making big changes in the AI industry by creating cost-effective language models that work just as well as those made by big companies like OpenAI and Meta. The company's main product, called R1, can think and reason as effectively as OpenAI's similar model. They have also released a new AI model called Janus Pro that can work with both text and images, and they say it works better than other popular tools like Stable Diffusion and DALL-E 3.

DeepSeek's chatbot, which competes with ChatGPT, became very popular very quickly and reached the top spot on the App Store. This success affected the stock market in an interesting way - shares of Nvidia, a major tech company, fell by 17 percent on January 27th, losing almost $600 billion in value. According to CNBC, this was the biggest one-day drop ever seen in US stock market history.

The company's AI helper uses their advanced DeepSeek-V3 model, which lets users do many things like asking questions, planning trips, and creating text. When too many people started downloading the app, DeepSeek had to limit new sign-ups because of what they called "malicious attacks" (harmful attacks) on their system.

Started in 2023 by Liang Wenfeng, DeepSeek has gotten attention because they can build open-source AI models using less money and fewer special computers compared to the billions spent by bigger companies like OpenAI, Meta, Google, and Microsoft. If DeepSeek is really performing as well as they say, it shows they can make powerful AI even though US rules stop companies like Nvidia from selling their best computer chips to China.

Here's the latest news about DeepSeek:

SEOUL, Feb 17 (Reuters) - South Korea's data privacy office announced on Monday that new downloads of the Chinese AI app DeepSeek have been stopped in the country. This happened after DeepSeek admitted they didn't follow all of the country's rules about protecting personal information.

The app will work again once they make changes to follow South Korea's privacy laws, said the Personal Information Protection Commission (PIPC) during a news meeting.

The rule, which started on Saturday, stops new people from downloading the app, but DeepSeek's website can still be used in South Korea.

The PIPC said that DeepSeek recently got legal representatives in South Korea and admitted they hadn't fully considered the country's data protection rules.

In related news, Italy's privacy office, the Garante, said last month they told DeepSeek to stop their chatbot in Italy because the company didn't address their concerns about privacy.

DeepSeek hasn't yet responded when asked to comment.

When asked about South Korea's actions to block DeepSeek, a spokesperson from China's foreign ministry spoke at a meeting on February 6. They explained that China takes data privacy and security very seriously and follows all laws to protect it.

The spokesperson also made it clear that China would never ask any company or person to collect or keep data in ways that break the law.

There's interesting news about Apple's AI plans in China. While Apple Intelligence is now available in the EU—and some say it's working in places where it wasn't supposed to—the company hasn't started offering its AI features in China yet. A new report from The Information on Tuesday suggests this might change soon. 

After looking at AI models from different Chinese companies like Tencent, ByteDance, Alibaba, and DeepSeek, Apple has asked Chinese officials to approve some features they developed with Alibaba.

Feb. 6, 2025 (WSJ) - DeepSeek faces similar challenges as TikTok. A new bill supported by both political parties wants to stop the Chinese AI chatbot DeepSeek from being used on devices owned by the US government. They say this is "prevent our enemy from getting information from our government." This reminds us of what happened with TikTok in 2020, when a similar ban was proposed, leading to its recent short shutdown and forced sale.

Several countries including Australia, Italy, and South Korea, along with the US state of Texas, have already put similar bans in place. The US Navy and NASA have also blocked the app within their organizations.

Feb 1. Nvidia announces its new GPUs work best with DeepSeek AI, but that might not be the main issue.

Nvidia is proud that DeepSeek's open source AI models work really well on their new RTX 50-series GPUs. They say these GPUs can “run the DeepSeek family of distilled models faster than anything on the PC market.”  However, this announcement might be missing something important.

This week, Nvidia lost more money in one day than any US company ever has, and many people say DeepSeek caused this. DeepSeek showed that their new R1 reasoning model doesn't need expensive Nvidia hardware to work as well as OpenAI's o1 model, which means they could train it for much less money. What DeepSeek did with R1 shows that maybe we don't absolutely need Nvidia's best chips to make progress in AI, which could affect Nvidia's success in the future.

Jan 31 AI uses lots of energy, but DeepSeek might change this.

Last month, DeepSeek surprised everyone by saying their AI model needs only about one-tenth of the computing power that Meta's Llama 3.1 model uses. This news changed how we think about the resources needed to develop AI.

This discovery could really affect how AI impacts our environment. Big tech companies are building huge AI data centers that might use as much electricity as small cities do. When we make this much electricity, it creates pollution. Many people worry that all the buildings and equipment needed for these new AI tools might make climate change worse and affect the air we breathe.

-----

Hashtags: #DeepSeekAI #TechRegulation

Source: theverge.com https://www.theverge.com/24353060/deepseek-ai-china-nvidia-openai