Until a number of days ago, only the nerdiest of nerds (I say that as such) had ever heard of it DeepSeeka Chinese AI subsidiary with the equally impressive name High-flyer capital managementa quantitative evaluation (or quant) company first founded in 2015.
Nevertheless, it has probably been essentially the most discussed company in Silicon Valley in recent days. This is basically due to the discharge of DeepSeek R1, a brand new large language model that performs similar “inference” to OpenAI's currently best available model o1 – it takes several seconds or minutes to reply difficult questions and solve complex problems because it goes thinks about his own evaluation step-by-step or “chain of thought”.
Additionally, DeepSeek R1 performed as well or higher than OpenAI's o1 on various third-party benchmarks (tests designed to measure AI performance in answering questions on various topics), and was reportedly even rated worse Training at a fraction of the price (reportedly around $5 million) with far fewer graphics processing units (GPU), under a strict embargo imposed by the US, OpenAI's home territory.
But unlike o1, which is simply available to paying ChatGPT subscribers on the Plus tier ($20 per 30 days) and costlier tiers (e.g. Pro at $200 per 30 days), DeepSeek R1 was launched as completely open source model released, which also explains why It has quickly catapulted up the charts of most downloaded and lively models on AI code-sharing community Hugging Face.
Additionally, due to the undeniable fact that it is totally open source, people have already refined and trained many alternative variations of the model for various task-specific purposes, akin to: B. to make it sufficiently small to run on a mobile device or to mix it with other open source models. Even if you should use it for development purposes, DeepSeek's API cost is greater than 90% cheaper than OpenAI's equivalent o1 model.
Most impressively, you don't even must be a software developer to make use of it: DeepSeek has one free website And mobile app also for US users with an R1-based chatbot interface very much like OpenAI's ChatGPT. Except once more: DeepSeek has undercut or “bullied” OpenAI by combining this powerful reasoning model with web search – something OpenAI has not done yet (web search is currently only available on the less powerful GPT family of models).
An open and closed irony
There's a slightly delicious, perhaps troubling, irony on this, given OpenAI's founding goals of democratizing AI for the masses. As Jim Fan, senior research manager at NVIDIA, brought it to X: “We live in a timeline where a non-US company is keeping OpenAI's original mission alive – truly open, groundbreaking research that empowers everyone.” It is unnecessary. The most entertaining consequence is the almost certainly.”
Or as X user @SuspendedRobot put itciting reports that DeepSeek appears to have been trained on question-answer output and other data generated by ChatGPT: “OpenAI stole your complete web to make itself richer, DeepSeek stole it and gave it back to the masses at no cost. I feel there’s a certain British tale about it.”
But Fan isn't the just one paying attention to DeepSeek's success. DeepSeek R1's open-source availability, its high performance, and the undeniable fact that it seemingly got here “out of nowhere” to challenge the previous leader in generative AI are, in keeping with my conversations and readings, sending shockwaves throughout Silicon Valley and much beyond also triggered various engineers, thinkers and executives. If “everyone” isn’t freaking out about it, as my hyperbolic headline suggests, it’s actually the talk of the town in tech and business circles.
A Message sent to Blindthe app for sharing anonymous gossip in Silicon Valley, has been making the rounds suggesting that Meta is in crisis due to the success of DeepSeek, which is Meta's own effort, the king of open source, with its Llama models -AI has quickly surpassed.
“That changes the entire game”
X user @tphuang wrote convincingly: “DeepSeek has commodified AI outside of the leading edge. An aha moment for me in the primary photo. “R1 is so less expensive than U.S. labor costs that many roles can be automated away in the following five years.” note later Why DeepSeek's R1 is more attractive to users than OpenAI's o1:
@tphaung also posted one compelling analogy as a matter: “Will DeepSeek be to LLM what Android became to the operating system world?”
Web entrepreneur Arnaud Bertrand also didn't mince his words concerning the amazing impact of DeepSeek's success. Write to X: “It can’t be emphasized enough how profoundly this changes your complete game. And not only by way of AI, additionally it is a large indictment of the US’s misguided try and halt China’s technological development, without which Deepseek may not have been possible (because the saying goes: necessity is the mother of all invention).”
The censorship problem
But others have made cautionary remarks about DeepSeek's rapid rise, arguing that as a startup operating out of China, the corporate is inevitably subject to that country's laws and content censorship requirements.
In fact, I've found this to be the case in my very own use of DeepSeek on the iOS app here within the US Do not answer questions on Tiananmen SquareScene of the pro-democracy student protests and uprisings of 1989 and the next violent crackdown by the Chinese military, This resulted in at the very least 200, possibly 1000’s, deathswhich gave him the nickname “Tiananmen Square massacrein Western media.
Ben Hylak, a former Apple human interface designer and co-founder of AI product analytics platform Dawn, posted on X how Asking about this topic caused DeepSeek R1 to get into an advanced loop.
As a representative of the press, I after all take freedom of opinion and expression very seriously and it might be one of the vital fundamental and undeniable causes that I support.
However, I could be remiss not to say that OpenAI's models and products, including ChatGPT, also refuse to reply a complete range of questions on even innocuous content – particularly those related to human sexuality and NSFW erotic/mature themes.
Of course, it's not an apples-to-apples comparison. And there can be some for whom reluctance to depend on foreign technology makes them skeptical of the last word value and utility of DeepSeek. But its performance and low price are undisputed.
And at a time when… 16.5% of all US goods are imported by ChinaI find it difficult to warn against using DeepSeek R1 resulting from censorship concerns or security risks – especially when the model code might be downloaded at no cost, taken offline, used on the device in secure environments, and refined at will.
I definitely recognize an existential crisis surrounding the “fall of the West” and the “rise of China” motivating among the full of life discussions surrounding DeepSeek, and others have already linked it to how US users have joined the app Xiaohongshu (also referred to as “Little Red Book”) When TikTok was briefly banned in that country, I used to be amazed at the standard of life in China portrayed within the videos shared there. The arrival of DeepSeek R1 comes inside this narrative context – one through which China appears to be on the rise (and clearly is by many measures) while the US appears to be in decline (and by many measures it also appears to be).
The first, but not the last, Chinese AI model to shake the world
Nor will or not it’s the last Chinese AI model to threaten the dominance of the Silicon Valley giants – even when, like OpenAI, they’re raising extra money than ever for his or her ambitions to develop artificial general intelligence (AGI), programs that humans at most surpass economically worthwhile work.
Just yesterday, one other Chinese model from TikTok parent company Bytedance called Doubao-1.5-pro – was released with performance comparable to OpenAI's non-reasonable GPT-4o model in third-party benchmarks, but again 1/50 of the price.
Chinese models have gotten so good so quickly that even those outside the tech industry are noticing: The magazine just published an article about DeepSeek's success and that of other Chinese AI efforts in addition to political commentator Matt Bruenig posted this on X: “I actually have been using Gemini, ChatGPT, and Claude extensively for NLRB document summarization for nearly a yr. Deepseek is best at this than anyone else. The chatbot version of that is free. The price for using the API is 99.5% lower than the worth of the OpenAI API. (shrug emoji)”
How does OpenAI react?
No wonder OpenAI co-founder and CEO Sam Altman Today the corporate said brought its yet-to-be-released second reasoning model family, o3, to ChatGPT at no cost users as well. OpenAI still appears to be forging its own path with more proprietary and advanced models, setting the industry standard.
But the query arises: With DeepSeek, ByteDance and other Chinese AI firms hot on its heels, how long can OpenAI remain on the forefront of developing and releasing latest, cutting-edge AI models? And if it falls, how strong and how briskly will its fall be?
However, there may be one other historical precedent for OpenAI. If DeepSeek and Chinese AI models actually turn into LLMs like Google's open source Android for mobile did – and took the lion's share of the marketplace for some time – just take a look at how the Apple iPhone does with its locked-down, proprietary All-in The House approach has managed to carve out the high-end segment of the market from there steadily expand downwardsparticularly within the US, to the purpose that the corporate now owns almost 60% of the domestic smartphone market.
Still, for anyone who spends a whole lot of money using AI models from leading labs, DeepSeek shows that the identical features could also be available less expensive and with much greater control. And in a company environment, that might be enough to win the sport.
