Browse all

What is DeepSeek?

The Chinese AI thats is causing a whole stir

What is DeepSeek? The Chinese AI thats is causing a whole stir

Since the introduction of artificial intelligence language models, OpenAI has undoubtedly managed to establish an almost monopolistic dominance over the vast majority of the public. In just over two years, ChatGPT has become an almost essential tool for various daily functions: some use it as a personal assistant, others for work, and some even as a therapist. Other tech giants have tried to replicate ChatGPT’s advanced intelligence, such as Google with Gemini, but with very modest resultsuntil last week. Like a bolt from the blue, during the first week of Donald Trump’s presidency and shortly after the failure of the TikTok ban in America, the open DeepSeek chatbot emerged from China — a new AI that is not only significantly cheaper than its Western counterparts but far more advanced than ChatGPT’s o1 model (which, until earlier this year, was considered the best on the market). But how did DeepSeek develop an AI of this scale so quickly? Unlike OpenAI, the Chinese start-up has made its latest AI system “open source”, sharing its underlying code with other companies and researchers. This approach allows third parties to develop and distribute their own products based on the same technologies, facilitating rapid adoption at lower costs. Within a week, the app surpassed ChatGPT as the most downloaded mobile application, thanks in part to the fact that its most advanced model, DeepSeek V3, is free to use for everyone.

@delvewithmel what’s the big deal about DeepSeek R1? - built an AI that rivals OpenAI's o1 at ~27x lower cost - proves pure reinforcement learning (no RLHF) can train top-tier reasoning models - R1 exhibits emergent reasoning capabilities (self-correction, long-chain thinking) - is open-source (weights available for research & dev) all achieved despite chip sanctions… proving that massive compute ≠ better AI #ai #deepseek #r1 #openai #o1 #chatgpt original sound - delvewithmel

The new AI didn’t just surprise with its innovative technology, but it also triggered a complete collapse of the international hardware market. According to the New York Times, the research paper from the team behind DeepSeek revealed that the Chinese start-up seemingly used only 2,000 Nvidia chips to train its AI. In comparison, most of the world’s leading tech companies typically use at least 16,000 chips. Fewer materials mean lower costs: according to the Hangzhou-based tech firm, the development of the V3 model required approximately $6 million, a figure significantly lower than the billions spent by competitors. This created a domino effect, culminating in a total collapse of the New York Stock Exchange. The Financial Times reported that Nvidia — the company behind the chips used to create most AIs — lost nearly $600 billion in stock value on Monday, January 27, marking the largest single-day drop in history. It wasn’t alone: Micron and Broadcom (both microchip manufacturers) also lost over 10% each. Even the major Silicon Valley players were not spared, with Alphabet, Microsoft, Meta, and Apple seeing stock declines between 2% and 4%. Nearly all trades closed in the red yesterday.

The rise of DeepSeek comes in a highly charged political climate, particularly between the United States and China, considering that just a week ago the U.S. Senate was seeking to ban TikTok over alleged ties to the Chinese government, accusing the app of sharing users' sensitive data with the Communist Party. On Monday, newly-elected President Donald Trump stated during a press conference that the emergence of DeepSeek should serve as a “wake-up call” for U.S. tech companies, adding that if the Chinese managed to develop such an advanced model while spending less, it’s «good news for us [USA].» Additionally, being a Chinese-developed model, many online users have noticed the AI’s cognitive bias, especially when questions relate to the Chinese Communist Party or modern events censored in China (such as the 1989 Tiananmen Square incident). Similarly, when asked about Taiwan, the AI responded by saying, «It is a historical and legal fact, as well as a general consensus of the international community, that Taiwan is an inalienable part of China. The Chinese government adheres to the One China principle and resolutely opposes any form of secessionist attempts or “Taiwan independence.”» However, it does not respond in the same way to other historically controversial figures or events. For example, regarding the Palestinian conflict, the AI replied, «As for the question of Palestine, China supports the just cause of the Palestinian people in restoring their legitimate national rights, the establishment of an independent Palestinian state through peaceful negotiations, and the two-state solution. It also calls on the international community to step up efforts to promote a quick resumption of peace talks between Palestine and Israel.»

DeepSeek-V3 thus positions itself as a watershed moment in the evolution of artificial intelligence, not only for its advanced technological capabilities but also for its economic and geopolitical impact. The model has demonstrated the ability to solve logical problems, answer questions, and write code in a manner competitive with the best systems on the market, as evidenced by standard benchmarks. However, despite its immediate success, some limitations remain. Experts have pointed out that the Chinese model was not designed to reason through complex problems, an area where OpenAI o3 seems more promising. OpenAI, however, has not yet made its new system accessible to the public, opting for a more cautious approach compared to DeepSeek's open-source strategy. According to the New York Times, DeepSeek's decision to go open-source has sparked significant controversy: on the one hand, it has accelerated innovation by enabling rapid adoption and customization of the model; on the other, it has reignited the debate over the security of such technologies. Many U.S. experts and lawmakers fear this openness could facilitate the spread of misinformation or pose other serious risks. Others, however, argue that limiting open-source technology in the United States could give China a strategic advantage, pushing American companies and researchers to develop systems based on Chinese technologies. The new ChatGPT model will need to be released before we can make further judgments.