Why everyone in AI is going crazy over Deepseek

A few days ago, only the nerdiest of nerds (I say this as one) had heard of Veterana Chinese AI subsidiary of the equally named High-flying capital managementa quantitative analysis firm (or Quant) that initially launched in 2015.

However, in recent days, it has arguably been the most discussed company in Silicon Valley. That’s largely thanks to the release of Deepseek R1, a new large language model that performs “reasoning” similar to the current O1 model available from OpenAI, taking multiple seconds or minutes to answer difficult questions and solve complex problems as it reflects your own analysis. in a step-by-step manner or “chain of thought.”

Not only that, but Deepseek R1 scored high or higher than Openai’s O1 in a variety of third-party benchmarks (tests to measure AI performance in answering questions on various topics), and it was reported. Trained at a fraction of the cost (reportedly around $5 million) with far fewer graphics processing units (GPUs) under a strict embargo imposed by the United States, OpenAi’s territory.

But unlike O1, which is available only to paying ChatGPT subscribers of the plus tier ($20 per month) and more expensive tiers (like Pro at $200 per month), Deepseek R1 was released as an open source model. , which also explains why The AI ​​Code Share Charts have quickly skyrocketed as the community embraces the most downloaded and active Face models..

Additionally, thanks to the fact that it is open source, people have already tuned and trained many multiple variations of the model for different task-specific purposes, such as making it small enough to run on a mobile device or combining it with other code models. open. Even if you want to use it for development purposes, Deepseek’s API costs are more than 90% cheaper than OpenAI’s equivalent O1 model.

Most impressive of all, you don’t even need to be a software engineer to use it: Deepseek has a free website and mobile application even for users with a ChatBot interface with R1 engine very similar to OpenAI’s CHATGPT. Except, once again, Deepseek underlined or “Mogged” OpenAi connecting this powerful reasoning model to web search – something OpenAi has not been done yet (web search is only available in the less powerful GPT model family today) .

An open and closed irony

There’s a rather delightful, or perhaps disconcerting, irony to this given Operai’s founding goals to democratize AI to the masses. As NVIDIA senior research manager Jim Fan put it on X: “We are living in a timeline in which a non-US company is keeping Operai’s original mission alive, truly open, frontier research that empowers everyone. It doesn’t make sense. The most entertaining outcome is the most likely.”

or like X user @SuspendedRobot Post itreference reports that Deepseek appears to have been trained In the question outputs of questions and other data generated by ChatGPT: “Operai stole from all over the Internet to get richer and deeper stole them and return it to the masses for free. I think there is a certain British folktale about this”

But Fan isn’t the only one to sit up and take note of Deepseek’s success. The open source availability of Deepseek R1, its high performance, and the fact that it seemingly “came out of nowhere” to challenge the former leader of generative AI, has sent shockwaves throughout Silicon Valley and far beyond, based on to my conversations and readings of Various engineers, thinkers and leaders. If it’s not “everyone” going crazy as my hyperbolic headline suggests, it’s certainly the talk of the town in tech and business circles.

TO Message posted in blindThe anonymous gossip sharing app in Silicon Valley has been doing the rounds suggesting that Meta is in crisis over Deepseek’s success due to how quickly it surpassed Meta’s efforts to be the king of open source AI with its models. CALLS.

‘This changes the whole game’

X user @tphuang wrote convincingly: “Deepseek has commoditized AI outside of the high-end. Lightbulb moment for me in the first photo. R1 is so much cheaper than US labor cost that many jobs will be automated in the next 5 years,” Later noticing Why Deepseek’s R1 is more attractive to users than even OpenAI’s O1:

“3 big problems with O1:
1) Too slow
2) Too expensive
3) Lack of end-user control/dependency on OpenAI
R1 solves them all. A company can buy its own NVIDIA GPUs, run these models. You don’t have to worry about additional costs or slow/unresponsive Operai servers”

@tphaung also raised a Compelling analogy as a question: “Will Deepseek LLM become Android in OS World?”

Web entrepreneur Arnaud Bertrand wasn’t done with words about the surprising implications of Deepseek’s success, either. writing in x: “There is no exaggeration how profoundly this changes the entire game. And not just with respect to AI, it is also a massive indictment of the United States’ misguided attempt to stop China’s technological development, without which Deepseek may not have been possible (as the saying goes, necessity is the mother of all things). inventions)”.

The issue of censorship

However, others have sounded warning notes about Deepseek’s rapid rise, arguing that as a startup operated by China, it is necessarily subject to that country’s content censorship laws and requirements.

In fact, my own use of Deepseek on the iOS app here in the US found that it would Do not answer questions about Tiananmen Squarethe site of the 1989 pro-democracy student protests and uprisings, and subsequent violent suppression by the Chinese military, resulting in at least 200, possibly thousands of deathsearning him the nickname”Tiananmen Square Massacre“In the Western media.

Ben Hylak, former Apple human interface designer and co-founder of product analytics platform AI Dawn, published in X How Asking about this topic made Deepseek R1 enter a tortuous circuit.

As a member of the press myself, of course, I take freedom of speech and expression extremely seriously and it is arguably one of the most fundamental and unarguable causes I champion.

However, we would be remiss not to note that OpenAI models and products, including chatgpt, also refuse to answer a wide range of questions about even innocuous content, especially pertaining to human sexuality and erotic/adult, NSFW subject matter.

It’s not an apples-to-apples comparison, of course. And there will be some for whom resistance to relying on foreign technology makes them skeptical of Deepseek’s ultimate value and utility. But its performance and low cost cannot be denied.

And at a time when 16.5% of all US goods are imported by ChinaIt’s hard for me to warn against using Deepseek R1 based on censorship concerns or security risks, especially when the model code is freely available to download, unplug, use on devices in secure environments, and for fine-tuning at will.

I definitely detect some existential crisis about the “fall of the West” and “rise of China”, motivating some of the lively discussion around Deepseek, and others have already connected it to HOw Us Users Joined Xiaohongshu App (also known as “Little Red Book”) When Tiktok was briefly banned in this country, only to be shocked at the quality of life in China depicted in videos shared there. The arrival of Deepseek R1 occurs in this narrative context, one in which China appears (and by many metrics is clearly) ascending, while the United States appears (and by many metrics, is also) declining.

The first but hardly the last Chinese AI model to shake the world

Nor will it be the last Chinese AI model to threaten the dominance of Silicon Valley giants, even as they, like OpenAi, raise more money than ever for their ambitions to develop artificial general intelligence (AGI), programs superior to humans’ economically valuable work. .

Yesterday, another Chinese model from Tiktok’s parent company Bytedance, called Doubao-1.5-pro -was released with performance matching OpenAi’s unreasonable GPT-4O model on third-party benchmarks, but again, at 1/50 the cost.

Chinese models have gotten so good, so fast, even those outside the tech industry are taking note: The economist The magazine just ran a piece on the success of Deepseek and that of other Chinese AI efforts and political commentator Matt Bruenig Posted in X that: “I have been extensively using Gemini, ChatGPT, and Claude for NLRB document summary for almost a year. Deepseek is better than all of them. The chatbot version is free. The price to use their API is 99.5% below the OpenAI API price. (shrug emoji)”

How does Operai respond?

Little Wonder, co-founder and CEO of Sam Altman today he said that the company It was bringing its second family of reasoning models, O3, yet to be released to chat even for free users. OpenAi still appears to be forging its own path with more proprietary and advanced models, setting the industry standard.

But the question is: with Deepseek, Bytedance and other Chinese AI companies nipping at their heels, how long can OpenAi stay ahead in making and releasing new cutting-edge AI models? And if so and when it falls, how difficult and how fast will its decline be?

However, OpenAi has another historical precedent for this. If Deepseek and Chinese AI models become LLMS like Google’s open source did on mobile, taking the lion’s share of the market share for a while, you just have to look at how Apple’s iPhone with its blocking, patented , patented, all-inclusive The house’s approach managed to carve the high end of the market and constantly expand downwards from thereespecially in the United States, to the point that it now holds almost 60% of the national smartphone market.

Still, for all big-dollar spenders using AI models from leading labs, Deepseek shows that the same capabilities can be available for much cheaper and much greater control. And in a business environment, that can be enough to win the ball game.

Daily insights on business use cases with VB daily

If you want to impress your boss, VB Daily has you covered. We give you the inside scoop on what companies are doing with generative AI, from regulatory changes to practical implementations, so you can share insights for maximum ROI.

Read our Privacy Policy

Thanks for subscribing. Check out more VB newsletters here.

An error occurred.