Why everyone in artificial intelligence is afraid of Debsik

As of a few days, only the most obsessive (I say this is one) DibsicIt is a Chinese subsidiary of artificial intelligence called equality High capital management,, The amount of analysis (or quantity) that was initially launched in 2015.

However, within the past few days, it can be said that they are the most discussed companies in Silicon Valley. This is largely thanks Deepseek R1 release, A new large language model leads to “thinking” similar to the current OPENAI model O1- Take seconds or multiple minutes to answer difficult questions and solve complex problems because it reflects its own analysis in step or “step by step, or” a series of thought “fashion.

Not only that, but DeepSeek R1 recorded higher or higher than Openai’s O1 on a variety of third -party standards (tests to measure artificial intelligence performance in answering questions on various topics), and it was said Train on a small part of the cost (is said about $ 5 million) With a much fewer graphics processing units (GPU) under a strict ban imposed by the United States, home grass in Openai.

But unlike O1, which is only available to pay the additional Chatgpt subscribers (20 dollars per month) and more expensive specifications (such as Pro at $ 200 per month), Deepseek R1 was released as a fully open source model, which also explains the cause as well He soon disturbed the plans to share the artificial intelligence code.

Also, thanks to the fact that it is completely open source, people have already set and trained many multiple differences in the model for different purposes for the task, such as making them small enough to run it on a mobile device, or combine it. Other open source models. Even if you want to use it for development purposes, Deepseek API costs are more than 90 % cheaper From OPENAI’s equivalent.

The most impressive ever, you don’t even need to be a software engineer to use: his Deepseek Free site and Mobile application even for users With a Chatbot interface that works with a R1 system is very similar to the Chatgpt from Openai. Except, again, it undermines Deepseek or “Mogged” Openai by connecting this powerful logical style to search on the web – something that Openai has not yet done (the web search is available only on the least powerful GPT family at the present time).

An open and closed paradox

There is a very delicious paradox, or perhaps a worrying paradox for this founding goals of Openai to give democracy to democracy to the masses. like Jim van, great research director at NVIDIA, put it on x: “We live in a schedule where a non-American company keeps the original task of Openai alive-with a truly open search, enables everyone. It does not make sense. The most entertaining result is the most likely.”

or X user @suspendrobot is its statusReference of reports that Dibsic appears to have been trained In the outputs of answering questions and other data created by Chatgpt: “Openai from the entire Internet to make itself richer, stole by Deepseek and returns them to the masses for free, I think there are some British peoples about this” “

But the fan is not the only one who sits and notes the success of Dibsic. The availability of the open source of Deepseek R1, his high performance, and the fact that he “did not come out of anywhere” to challenge the former Ai leader, sent a shock throughout the Silicon Valley, and based on my talks and readings. Various engineers, thinkers and leaders. If “everyone” does not scare this as my excessive address suggests, then it is certain that the city talks in technology and business circles.

A The message published to the blindThe application of the unknown gossip in the Silicon Valley makes tours indicating that the death is in a crisis due to the success of Deepseek due to the speed that exceeded its Meta efforts to be the AI ​​king open source with Llama models.

“This changes the entire game.”

X user TPHUANG was convincingly: “Dibsic has collected artificial intelligence outside the very high -end. Later note Why is Deepseek’s R1 more tempting for users than Openai’s O1:

3 huge issues w/ O1:
1) Very slow
2) Very expensive
3) Failure to control the end user/rely on Openai
R1 solves each of them. The company can buy its NVIDIA GPUS, and run these models. Do not worry about additional costs or slow/non -respondable Openai servers “

Tphaung has also been released Casting measurement as a question: “Will Deepseek to LLM be what has become Android to the world of OS?”

Businessman on the Web Arnoud Bartrand did not produce words about the amazing effects of Dibsic’s success, either, Writing on x: “There is no exaggeration in the extent of changing this entire game. Not only with regard to Amnesty International, it is also an enormous indictment for the US misleading attempt to stop the technological development of China, without which Deepsek was not possible (as the proverb says, the necessity is the mother of inventions ).

The issue of censorship

However, others have seen warning notes about Deepseek, on the pretext that as a startup operating outside China, it is necessarily subject to the laws of that country and content control requirements.

In fact, I found my own use of Deepseek on the iOS app in the United States that it would Do not answer questions about Tiananmen SquareThe site of protests for students supporting democracy for the year 1989, the uprising, and the subsequent violent campaign by the Chinese army, Which led to at least 200, perhaps thousands of deathsWin it the title “Tiananmen Square massacre“In the Western media.

Ben Heilak, a human interface designer for the former Apple company and co -founder of the AI ​​DAWN product analysis platform, was published on X How The question about this topic caused Deepseek R1 to enter.

As a member of the press itself, I of course bring freedom of expression and expression seriously, and it can be said that it is one of the most basic reasons that cannot be a hero.

However, I will be intended not to point out that Openai’s models and products, including Chatgpt, also refuses to answer a full set of questions about harmless content – especially with regard to human race and sexual/adult subject, NSFW theme.

It is not a comparison of apples, of course. There will be some who make resistance to relying on foreign technology skeptical of the final value and benefit of Dempsic. But there is no denial of his performance and low cost.

And at a time when 16.5 % of all American goods are imported by ChinaIt is difficult for me to warn against using Deepseek R1 based on control concerns or security risks-especially when the model icon is available for free to download, enjoy the Internet, use the device in safe environments, and to determine the will. .

Certainly I discovered some existential crisis about “the fall of the West” and “the rise of China”, which stimulates some moving discussions about Deepseek, however, others have already linked it to H.OW Us Users join the Xiaohongshu app (also known as “Little Red Book”) When Tiktok was briefly banned in this country, just for the surprise of the quality of life in China shown in joint videos there. Deepsek R1 arrival occurs in this narrative context – in which China appears (and through many scales clearly) ascending while the United States (and on many standards, too) decrease.

The first Chinese Chinese model shaking the world

Also, it will not be the last Chinese model of artificial intelligence to threaten the dominance of the giants of the Silicon Valley – even because, like Openai, Upon more money from ever For their aspirations for the development of artificial general intelligence (AGI), programs that surpass humans in most businesses of economic value.

Only yesterday, another Chinese model is called the mother Tiktok company – Doubao-1.5-PRO -It was released with the Openi GPT-4O Again, at 1/10 cost.

Chinese models have become very good and very fast, even those who go beyond the technology industry live: Economic The magazine was just running an article on Deepseek’s success And the efforts of other Chinese Amnesty International and the political commentator Matt Proinalge posted on X: “I was using Gemini, Chatgpt and Claude widely for a summary of the NLRB document for about a year. Dempsic is better than each of them. Chatbot version is free. API use price is 99.5 % less than API Openai. (Ignore emojis) “

How does Openai respond?

Little Wonder Openai co -founder and CEO Sam Al -Taman Today he said that the company It was bringing the second Thinking Family that was released yet, O3, to ChatGPT even to free users. Openai still seems to descend its own path with more special and advanced models – which puts the industry standard.

But the question becomes: with Deepseek, bytedance, and other Chinese artificial intelligence companies who accept it, for how long can Openai can remain in the introduction and issue new models of artificial intelligence? And if that and when it falls, how difficult it is to decrease?

Openai has another historical precedent. If Deepseek and Chinese AI models are already to LLMS as an open source Android did from Google to Mobile-take the lion’s share from the market for a while-you just have to see how Apple iPhone with its lock, ownership, everyone in the house approach can sculpt from The high -end market and Expanding steadily from thereEspecially in the United States, to the point that it now possesses approximately 60 % of the local smartphone market.

However, for all those who spend a great Paczes to use artificial intelligence models from leading laboratories, Deepseek shows that the same capabilities may be available to much cheaper and with much larger control. In preparing the institution, this may be sufficient to win the game.

Daily visions about business use cases with VB daily

If you want to persuade your boss at work, you have covered VB Daily. We give you the internal journalistic precedence over what companies do with obstetric artificial intelligence, from organizational transformations to practical publishing operations, so that you can share visions of the maximum return on investment.

Read with us privacy policy

Thanks for subscribing. Check more VB news bulletins here.

An error occurred.