Why everyone in the AI ​​flips out via Deepseek

From a few days ago, only the nerdiest of nerds (I say that as one) had ever heard DeepseekA Chinese AI High-flyer capital managementA quantitative analysis (or quantitative) company that was initially introduced in 2015.

In the past few days it has probably been the most visited company in Silicon Valley. This is largely thanks Release from Deepseek R1, A new large voice model that resembles “argument” like the current best available model O1 from Openai-durably for several seconds or minutes to answer difficult questions and solve complex problems because it reflects its own analysis in one step by step, or “chain of thoughts” fashion.

Not only that, but also Deepseek R1 achieved a variety of third -party benchmarks (tests for measuring the AI trains at a fraction of the costs (reports around 5 million US dollars) With far fewer graphics processing units (GPU) under a strict embargo that the United States imposed, Openai’s Home lawn.

But in contrast to O1, which is only available for the payment of Chatgpt subscribers of the Plus -Tier (20 USD per month) and more expensive levels (such as Pro for $ 200 per month), Deepseek R1 was as a completely open source – Model published, which also explains why why It quickly set up the charts of the most downloaded and active models from AI Code Sharing Community -Sharing Community.

Thanks to the fact that it is completely open source, people already have many multiple variations of the model for different tasks, as small enough to run on a mobile device. Other open source models. Even if you want to use it for developmental purposes, The API costs of Deepseek are more than 90% cheaper As the equivalent O1 model from Openaai.

The most impressive of all is that you don’t even have to be a software engineer to use it: Deepseek has one Free website And Mobile app even for us users With a R1-poured chat bot interface that is very similar to the Openai chat. Apart from the fact that Deepseek Openaai or “Mogged” Openai connects to the web search by connecting this powerful argumentation model – some Openaai has not yet carried out (the web search is only available for the less powerful GPT family of models).

An open and closed irony

There is a fairly delicious or perhaps worrying irony that Openai’s founding goals for democratization of the AI ​​to the masses. As Nvidia Senior Research Manager Jim Fan put it on X: “We live in a timeline in which a non-US company live the original mission of Openaai-realized, open, border research, all of which enables. It makes no sense. The most entertaining result is the most likely. “

Or how X user @supendedrobot expose itReference reports that Deepseek seems to have been trained In question-and-work outputs and other data generated by Chatgpt: “Openai stolen from the entire Internet to make yourself richer, Deepseek stole them and give it back to the masses for free.

But fan is not the only one who looks up and that Deepseek’s success take note of. The Open Source availability of Deepseek R1, its high performance and the fact that it seemed to be out of nowhere to challenge the former leader of the generative AI different engineers, thinkers and managers. If not “everyone” flips out about how my hyperbolic heading suggests, it is certainly the city’s conversation in technology and business circles.

A Message posted blindThe app for parts of anonymous gossip in Silicon Valley has made the round that indicates that Meta is in the crisis about Deepseek’s success because it is the king of open source AI with his Llama models to be exceeded.

“This changes the whole game”

X user @Tphuang wrote mandatory: “Deepseek awarded the AI ​​outside the very top end. Light bulb for me on the 1st photo. R1 is as cheaper than the US work costs that many jobs will be automated over the next 5 years. ” Later noticed Why Deepseeks R1 is more tempting for users than even Openas O1:

“3 big problems with O1:
1) too slow
2) too expensive
3) Lack of control for end users/trust in Openai
R1 solves all. A company can buy its own Nvidia GPUS and run these models. You don’t have to worry about additional costs or slow/non -reacting Openai servers. “

@Tphaung also set a convincing analogy as a question: “Will Deepseek be too LLM, which Android became OS World?”

The web entrepreneur Arnaud Bertrand also broke no words about the astonishing effects of Deepseek’s success. Writing on x: “There is not too overvalued how deeply this changes the entire game. And not only in relation to AI, it is also a massive indictment against the failed attempt by the United States to stop China’s technological development without which Deepseek may not be possible (as the saying says, the need for the mother of inventions). “

The censorship edition

However, others have set up as a precautionary level for Deepseek’s quick ascent and argued that it is necessarily subject to the laws and censorship requirements of this country as a startup that was shown from China.

In fact, I found my own use of Deepseek in the iOS app here in the USA Do not answer any questions about the Tiananmen SquareThe location of the protests of the protests and the uprising of the protests in 1989 and then violent procedure of the Chinese military, Which leads to at least 200, possibly thousands of deathsearn the nickname “Massaker on the Tiananm Square”In western media.

Ben Hylak, a former designer and co -founder of the AI ​​Product Analytics Platform Dawn, has on x how If you ask about this topic, Deepseek R1 entered a cumbersome loop.

As a member of the press itself, I naturally take the freedom of expression and expression very seriously and it is probably one of the most fundamental and untembling causes that I have champion.

However, I would not find that the models and products from Openaai, including chatt, also refuse to answer a whole series of questions about self -harmless content – especially in connection with human sexuality and erotic/adult, NSFW issues.

Of course, it is not a apples-to-color comparison. And there will be some for which the resistance to foreign technology makes you skeptical and benefit from Deepseek’s final value. But it cannot be denied that its performance and low costs are favorable.

And at a time when 16.5% of all US goods are imported by ChinaIt is difficult for me to use the use of Deepseek R1 based on censorship problems or security risks-especially if the model code freely download, offline, in safe environments in secure environments and has been tuned and coordinated at will. .

I definitely recognize an existential crisis about the “fall of the West” and the “rise of China”, which motivates some of the animated discussions about Deepseek, and others have already connected them to HOW US users have joined the Xiaohongshu app (also known as “Little Red Book”). When Tikkok was briefly banned in this country to amazing the quality of life in China in the videos shared there. The arrival of Deepseek R1 occurs in this narrative context in which China appears (and after many metrics is clear), while the USA (and after many metrics) appears in decline.

The first, but hardly the last Chinese KI model to shake the world

It will not be the last Chinese AI model that endangers the dominance of Silicon Valley Giants – even if you like Openaai, collect more money than ever For their ambitions to develop artificial general intelligence (AGI), exceed programs that exceed humans at most.

Just yesterday another Chinese model of the TIKTOK mother company bytedance – called Doubao-1.5-Pro -wurde with the non-pending GPT 4O model from Performance Matching Openai for benchmarks published by third-party providers, but published Again at 1/50 the costs.

Chinese models have become so good, so quickly, even those outside the tech industry take note of it: The economist The magazine has just carried out a piece via Deepseek’s success and the other Chinese AI efforts and political commentator Matt bruenig posted the: “I have used Gemini, Chatgpt and Claude extensively for the summary of the NLRB document for almost a year. Deepseek is better than everyone. The Chatbot version is free of charge. The price for using the API is 99.5% below the price of the Openai -API. (Shrug emoji) ”

How does Openai react?

No wonder Openai co-founder and CEO Sam Altman Today the company said I brought his second argumentation model family, O3, who has not yet been published, also for free users in chatgpt. Openaai still seems to be on its own way with more proprietary and more advanced models – which defines the industrial standard.

But the question becomes: With Deepseek, Bytedance and other Chinese AI companies that flee on the heels, how long can you open when you manufacture and release new state-of-the-art AI models? And if it and if it falls, how difficult and how fast will his decline be?

However, Openaai has another historical precedent. When Deepseek and Chinese AI models actually become LLMs, such as Google’s Open Source Android for Mobile Devices for a while-the lion’s share of the market takes over the only way to see the house approach to cut off and the high end of the market extend steadily from thereEspecially in the United States, it now has almost 60% of the domestic smartphone market.

Nevertheless, Deepseek shows for all these expenses for a lot of money to use AI models from leading laboratories, for much cheaper and with much greater control the same functions. And in an Enterprise environment, this can be enough to win the ball game.