DeepSeek continues to be showered with reward from tech trade executives and lawmakers alike, days after the Chinese language AI startup launched its reasoning mannequin R1, which triggered a broader sell-off in tech shares throughout markets from New York to Tokyo.
The standard and cost-efficiency of its R1 mannequin is what has primarily led to DeepSeek’s surge in reputation. The corporate has claimed that its AI mannequin matches with, and in some instances beats, OpenAI’s o1 reasoning mannequin in efficiency whereas utilizing fewer graphics processing models (GPUs) and costing far much less.
DeepSeek’s chatbot app, which offers free entry to R1, has risen to the highest of app retailer charts in a number of international locations. Nevertheless, the corporate’s success story has been met with warning and skepticism by some. OpenAI has accused DeepSeek of IP theft and mentioned it has proof of the corporate utilizing its GPT fashions to coach its personal.
For sure, the thrill round DeepSeek is giving strategy to scrutiny with questions rising about how its AI fashions have been developed and varied interpretations about their broader impression. Because the dialogue unfolds, let’s break down a few of the widespread myths surrounding DeepSeek’s rise.
- 01
Fable #1: DeepSeek’s AI fashions sign AGI is inside attain
Actuality: DeepSeek’s AI fashions are a major enchancment in effectivity and value, however they don’t essentially point out a leap in direction of synthetic normal intelligence (AGI).
AGI is a time period utilized by the tech trade to explain an AI mannequin able to equaling or surpassing human mind on a variety of duties. Nobody has declared that they’ve developed such an AI mannequin but. Nevertheless, OpenAI and a few of its rivals have mentioned that they’re eagerly working in direction of reaching the AGI milestone.
In 2023, DeepSeek reportedly developed from the AI analysis unit of a Chinese language hedge fund, Excessive-Flyer, to an AI firm. The agency was established by hedge fund supervisor Liang Wenfeng with the target of growing massive language fashions (LLMs) on the trail to AGI.
OpenAI CEO Sam Altman has repeatedly expressed confidence that the ChatGPT-maker will obtain AGI. In his response to the thrill round DeepSeek, Altman once more shifted focus to AGI whereas hailing the R1 mannequin as ‘spectacular’.
Though R1 marks an inflection level within the race for AI supremacy, DeepSeek has not launched solely new expertise. “Attending to AGI most likely requires 5 or 6 extra breakthroughs and the corporate or nation that may ramp up these breakthroughs first might win,” Gary Marcus, New York College (NYU) professor and AI knowledgeable, advised CNBC.
- 02
Fable #2: DeepSeek’s breakthrough exhibits export controls don’t work
Actuality: US export restrictions on the sale of superior GPUs might proceed to have a major impression on China’s AI growth.
DeepSeek’s breakthrough has been seen because the unintended end result of US export controls that restricted Chinese language tech companies from buying superior GPUs to scale their AI fashions. With out entry to Nvidia’s top-of-the-line chips, DeepSeek researchers have been reportedly pressured to give you intelligent methods to make AI fashions extra environment friendly of their consumption of uncooked compute energy.
Critics have argued that US export controls backfired, however DeepSeek reportedly stockpiled 10,000 of Nvidia’s older era A100 GPUs earlier than the commerce restrictions have been imposed.
Miles Brundage, an AI coverage knowledgeable who lately left OpenAI, has urged that export controls may nonetheless sluggish China down in terms of working extra AI experiments and constructing AI brokers.
“DeepSeek was pressured by means of necessity to search out a few of these methods possibly sooner than American firms may need. However that doesn’t imply they wouldn’t profit from having far more. That doesn’t imply they’re able to instantly leap from o1 to o3 or o5 the way in which OpenAI was in a position to do, as a result of they’ve a a lot bigger fleet of chips,” Brundage mentioned in a current podcast interview.
As well as, Dario Amodei, the CEO of Anthropic, the corporate behind the Claude sequence of AI fashions, has mentioned that DeepSeek’s outcomes “make export management insurance policies much more existentially necessary than they have been every week in the past.”
- 03
Fable #3: DeepSeek is a grave menace to Nvidia
Actuality: DeepSeek’s R1 mannequin will not be as regarding for Nvidia as some may assume.
The excitement round DeepSeek brought on panic amongst Nvidia buyers, which resulted in its shares dipping by 17 per cent and wiping out practically $600 billion in market worth on January 27. The chip big’s inventory recovered from the sharp hunch on January 28, it fell one other 4 per cent on January 29.
Whereas DeepSeek’s R1 mannequin might have diminished the requirement of huge arrays of particular goal AI {hardware} from the likes of Nvidia, it doesn’t precisely spell doom for the chip big.
Microsoft CEO Satya Nadella identified that DeepSeek’s impression may, counterintuitively, improve demand for superior GPUs. “Jevons paradox strikes once more!” Nadella wrote in a put up on X.
Jevons Paradox is an financial concept which means that when technological progress makes using a useful resource extra environment friendly, total consumption of that useful resource tends to extend.
Tech investor Andrew Ng additionally mentioned that it stays to be seen if DeepSeek’s outcomes will cut back the demand for GPUs and compute energy. “Generally making every unit of cheaper can lead to extra {dollars} in complete going to purchase that good,” he mentioned in a put up on X.
- 04
Fable #4: DeepSeek R1 is a totally open-source mannequin
Actuality: DeepSeek R1 will be downloaded, modified, and reused without cost, but it surely will not be thought-about really open supply.
The spectacular outcomes of DeepSeek R1 has been interpreted by many as an indication of China pulling forward of the US within the race for AI supremacy. However past the geopolitical angle, DeepSeek’s success can be being celebrated as a win of open-source AI over closed AI.
Echoing this sentiment, Meta’s chief AI scientist, Yann LeCun, mentioned, “DeepSeek has profited from open analysis and open supply (e.g., PyTorch and Llama from Meta). They got here up with new concepts and constructed them on high of different individuals’s work. As a result of their work is revealed and open supply, everybody can revenue from it. That’s the energy of open analysis and open supply.”
R1’s underlying mannequin structure and weights (numerical values used to point how an AI mannequin processes info) has been made publicly accessible underneath a permissive MIT licence. Which means the mannequin will be deployed with out restrictions.
However R1 doesn’t match the broadly accepted definition of ‘open-source’. In line with the Open Supply Initiative (OSI), a really open-source AI mannequin should present entry to particulars in regards to the knowledge used to coach the AI, the entire code used to construct and run the AI, and the settings and weights from the coaching.
The info used to coach R1 has not been made accessible. The coaching code and different directions for coaching haven’t been supplied both. Open-source AI builders are typically cautious of releasing coaching datasets as it might invite copyright infringement lawsuits.
- 05
Fable #5: DeepSeek’s AI fashions carry further privateness threat
Actuality: DeepSeek’s AI poses the identical threat to privateness as different LLMs
DeepSeek’s meteoric rise has been accompanied by knowledge privateness issues amongst customers and authorities. A few of these issues have been fueled by the AI startup’s Chinese language origins whereas others have pointed to the open-source nature of its AI expertise.
In its privateness coverage, DeepSeek unequivocally states: “We retailer the knowledge we acquire in safe servers positioned within the Folks’s Republic of China.”
Nevertheless, tech trade figures akin to Perplexity CEO Aravind Srinivas have repeatedly sought to allay such worries by stating that DeepSeek’s R1 mannequin will be downloaded and run regionally in your laptop computer or different units. Working native situations implies that customers can privately work together with DeepSeek’s AI with out the corporate getting their palms on enter knowledge.
In line with Srinivas, Perplexity is internet hosting the R1 mannequin in knowledge centres positioned within the US and European Union (EU), not China. He additionally claimed that the Perplexity-hosted model of R1 is free from censorship restrictions.
© The Indian Categorical Pvt Ltd
