DeepSeek decoded: 5 myths and realities about the Chinese AI startup’s rise

DeepSeek’s chatbot app, which provides escaped entree to R1, has risen to the apical of app store charts successful aggregate countries. However, the company’s occurrence communicative has been met with caution and skepticism by some. OpenAI has accused DeepSeek of IP theft and said it has grounds of the institution utilizing its GPT models to bid its own.

Needless to say, the excitement astir DeepSeek is giving mode to scrutiny with questions emerging astir however its AI models were developed and assorted interpretations astir their broader impact. As the treatment unfolds, let’s interruption down immoderate of the communal myths surrounding DeepSeek’s rise.

1. 01

   Myth #1: DeepSeek’s AI models awesome AGI is wrong reach

   Reality: DeepSeek’s AI models are a important betterment successful ratio and cost, but they don’t needfully bespeak a leap towards artificial wide quality (AGI).

   AGI is simply a word utilized by the tech manufacture to picture an AI exemplary susceptible of equaling oregon surpassing quality intellect connected a wide scope of tasks. No 1 has declared that they person developed specified an AI exemplary yet. However, OpenAI and immoderate of its rivals person said that they are eagerly moving towards reaching the AGI milestone.

   In 2023, DeepSeek reportedly evolved from the AI probe portion of a Chinese hedge fund, High-Flyer, to an AI company. The steadfast was established by hedge money manager Liang Wenfeng with the nonsubjective of processing ample connection models (LLMs) connected the way to AGI.

   OpenAI CEO Sam Altman has repeatedly expressed assurance that the ChatGPT-maker volition execute AGI. In his effect to the buzz astir DeepSeek, Altman again shifted absorption to AGI portion hailing the R1 exemplary arsenic ‘impressive’.

   Although R1 marks an inflection constituent successful the contention for AI supremacy, DeepSeek has not introduced wholly caller technology. “Getting to AGI astir apt requires 5 oregon six much breakthroughs and the institution oregon state that tin ramp up those breakthroughs archetypal whitethorn win,” Gary Marcus, New York University (NYU) prof and AI expert, told CNBC.

2. 02

   Myth #2: DeepSeek’s breakthrough shows export controls don’t work

   Reality: US export restrictions connected the merchantability of precocious GPUs whitethorn proceed to person a important interaction connected China’s AI development.

   DeepSeek’s breakthrough has been viewed arsenic the unintended result of US export controls that restricted Chinese tech firms from purchasing precocious GPUs to standard their AI models. Without entree to Nvidia’s top-of-the-line chips, DeepSeek researchers were reportedly forced to travel up with clever ways to marque AI models much businesslike successful their depletion of earthy compute power.

   Critics person argued that US export controls backfired, but DeepSeek reportedly stockpiled 10,000 of Nvidia’s older procreation A100 GPUs earlier the commercialized restrictions were imposed.

   Miles Brundage, an AI argumentation adept who precocious near OpenAI, has suggested that export controls mightiness inactive dilatory China down erstwhile it comes to moving much AI experiments and gathering AI agents.

   “DeepSeek was forced done necessity to find immoderate of those techniques possibly faster than American companies mightiness have. But that doesn’t mean they wouldn’t payment from having overmuch more. That doesn’t mean they are capable to instantly leap from o1 to o3 oregon o5 the mode OpenAI was capable to do, due to the fact that they person a overmuch larger fleet of chips,” Brundage said successful a caller podcast interview.

   In addition, Dario Amodei, the CEO of Anthropic, the institution down the Claude bid of AI models, has said that DeepSeek’s results “make export power policies adjacent much existentially important than they were a week ago.”

3. 03

   Myth #3: DeepSeek is simply a sedate menace to Nvidia

   Reality: DeepSeek’s R1 exemplary whitethorn not beryllium arsenic concerning for Nvidia arsenic immoderate mightiness think.

   The buzz astir DeepSeek caused panic among Nvidia investors, which resulted successful its shares dipping by 17 per cent and wiping retired astir $600 cardinal successful marketplace worth connected January 27. The spot giant’s banal recovered from the crisp slump connected January 28, it fell different 4 per cent connected January 29.

   While DeepSeek’s R1 exemplary whitethorn person diminished the request of immense arrays of peculiar intent AI hardware from the likes of Nvidia, it does not precisely spell doom for the spot giant.

   Microsoft CEO Satya Nadella pointed retired that DeepSeek’s interaction could, counterintuitively, summation request for precocious GPUs. “Jevons paradox strikes again!” Nadella wrote successful a station connected X.

   Jevons Paradox is an economical mentation which suggests that erstwhile technological advancement makes the usage of a assets much efficient, wide depletion of that assets tends to increase.

   Tech capitalist Andrew Ng besides said that it remains to beryllium seen if DeepSeek’s results volition trim the request for GPUs and compute power. “Sometimes making each portion of a bully cheaper tin effect successful much dollars successful full going to bargain that good,” helium said successful a station connected X.

4. 04

   Myth #4: DeepSeek R1 is simply a afloat open-source model

   Reality: DeepSeek R1 tin beryllium downloaded, modified, and reused for free, but it whitethorn not beryllium considered genuinely unfastened source.

   The awesome results of DeepSeek R1 has been interpreted by galore arsenic a motion of China pulling up of the US successful the contention for AI supremacy. But beyond the geopolitical angle, DeepSeek’s occurrence is besides being celebrated arsenic a triumph of open-source AI implicit closed AI.

   Echoing this sentiment, Meta’s main AI scientist, Yann LeCun, said, “DeepSeek has profited from unfastened probe and unfastened root (e.g., PyTorch and Llama from Meta). They came up with caller ideas and built them connected apical of different people’s work. Because their enactment is published and unfastened source, everyone tin nett from it. That is the powerfulness of unfastened probe and unfastened source.”

   R1’s underlying exemplary architecture and weights (numerical values utilized to bespeak however an AI exemplary processes information) has been made publically disposable nether a permissive MIT licence. This means that the exemplary tin beryllium deployed without restrictions.

   But R1 doesn’t acceptable the wide accepted explanation of ‘open-source’. According to the Open Source Initiative (OSI), a genuinely open-source AI exemplary indispensable supply entree to details astir the information utilized to bid the AI, the implicit codification utilized to physique and tally the AI, and the settings and weights from the training.

   The information utilized to bid R1 has not been made available. The grooming codification and different instructions for grooming person not been provided either. Open-source AI developers are mostly wary of releasing grooming datasets arsenic it whitethorn invitation copyright infringement lawsuits.

5. 05

   Myth #5: DeepSeek’s AI models transportation other privateness risk

   Reality: DeepSeek’s AI poses the aforesaid hazard to privateness arsenic different LLMs

   DeepSeek’s meteoric emergence has been accompanied by information privateness concerns among users and authorities. Some of these concerns person been fueled by the AI startup’s Chinese origins portion others person pointed to the open-source quality of its AI technology.

   In its privateness policy, DeepSeek unequivocally states: “We store the accusation we cod successful unafraid servers located successful the People’s Republic of China.”

   However, tech manufacture figures specified arsenic Perplexity CEO Aravind Srinivas person repeatedly sought to allay specified worries by pointing retired that DeepSeek’s R1 exemplary tin beryllium downloaded and tally locally connected your laptop oregon different devices. Running section instances means that users tin privately interact with DeepSeek’s AI without the institution getting their hands connected input data.

   According to Srinivas, Perplexity is hosting the R1 exemplary successful information centres located successful the US and European Union (EU), not China. He besides claimed that the Perplexity-hosted mentation of R1 is escaped from censorship restrictions.

© The Indian Express Pvt Ltd