UPSC Issue at a Glance | DeepSeek Breakthrough: 4 Key Questions You Must Know for Prelims and Mains

What is the issue?

In the past fewer days, the satellite has turned upside down. Tech stocks person mislaid $1 trillion, and the United States is nary longer the sole person of artificial quality (AI), arsenic it erstwhile claimed. This displacement successful dynamics is each due to the fact that of the meteoric emergence of Chinese AI startup DeepSeek. In caller weeks, DeepSeek has captured planetary attraction and shaken up Silicon Valley and Washington, D.C., with the instauration of its AI models—DeepSeek-V3 and DeepSeek-R1, a reasoning model. In what immoderate are calling a “Sputnik moment,” DeepSeek appears to person surpassed companies similar OpenAI, Google, and Meta successful the high-stakes AI race.

DeepSeek breakthrough is simply a nonstop situation to the thought that AI advancement depends connected tremendous computational power, immense datasets, and billions successful backing (Reuters illustration)

Why is this contented relevant?

The DeepSeek breakthrough is applicable for the UPSC CSE exam due to the fact that artificial quality and emerging technologies are integral topics successful General Studies Paper III. Furthermore, UPSC has antecedently asked questions connected AI, frankincense knowing astir the large developments successful this domain is crucial. UPSC aspirants volition besides find this taxable utile for essays and existent affairs, arsenic good arsenic for their property tests.

UPSC Syllabus: 

Preliminary Examination: Current events of nationalist and planetary importance, Science and Technology

Mains Examination: General Studies-II,III: Awareness successful the fields of IT, Space, Computers, robotics, nano-technology, bio-technology and issues relating to intelligence spot rights, Government policies and interventions 

Question 1: What is DeepSeek? 

DeepSeek is simply a Chinese AI company located successful Hangzhou, founded by entrepreneur Liang Wenfeng, who besides serves arsenic the CEO of the quantitative hedge money High Flyer. Wenfeng began moving connected AI successful 2019 with his company, High Flyer AI, which focuses connected probe successful this field. 

Recently, DeepSeek launched its AI models—DeepSeek-V3 and DeepSeek-R1, a reasoning model. These models rapidly gained popularity, surpassing ChatGPT to go the astir downloaded app connected the App Store. DeepSeek-V3 and DeepSeek-R1 vie with OpenAI’s precocious models, o1 and o3, as the Chinese laboratory achieved this feat lone with a fraction of their investments.

Here are a fewer different open-source AI models developed by DeepSeek:

📍DeepSeek Coder: An open-source AI exemplary designed for coding-related tasks.

📍DeepSeek LLM: An AI exemplary with a 67 cardinal parameter number to rival different ample connection models (LLMs).

📍DeepSeek-V2: A low-cost AI exemplary that boasts of beardown performance.

📍DeepSeek-Coder-V2: An AI exemplary with 236 cardinal parameters designed for analyzable coding challenges.

📍DeepSeek-V3: A 671 cardinal parameter AI exemplary that tin grip a scope of tasks specified arsenic coding, translating, and penning essays and emails.

📍DeepSeek-R1: An AI exemplary designed for reasoning tasks, with capabilities that situation OpenAI’s marquee o1 model.

📍DeepSeek-R1-Distill: An AI exemplary that has been fine-tuned based connected synthetic information generated by DeepSeek R1.

In this context, a question people arises: What makes DeepSeek AI models unique, and however bash they basal isolated from different AI players? Let’s recognize what sets DeepSeek isolated successful the evolving satellite of artificial intelligence!

Question 2: How is DeepSeek antithetic from different AI players? 

DeepSeek appears to person surpassed large players similar OpenAI, Google, and Meta successful the competitory scenery of AI development. The lab’s recently released open-source reasoning model, DeepSeek R1, is reported to outperform starring AI models, specified arsenic OpenAI’s o1, connected cardinal mathematics and reasoning benchmarks. Broadly, two factors marque DeepSeek the speech of the municipality .i.e. its state-of-the-art exertion and affordable cost.

State-of-the-art technology

1. Open-sourced nature: DeepSeek models are open-source, dissimilar the closed models from OpenAI and Google. This means that different companies, particularly tiny developers, tin physique connected apical of DeepSeek’s exemplary and amended it without paying licence fees. The imaginable is huge—rather than processing their ain models, companies tin modify and deploy DeepSeek’s models astatine a fraction of the cost. This could thrust wide AI adoption astatine scale.

Bijin Jose writes– What sets DeepSeek models isolated is their show and open-sourced quality with unfastened weights, which fundamentally allows anyone to physique connected apical of them. The DeepSeek-V3 has been trained connected a meagre $5 million, which is simply a fraction of the hundreds of millions pumped successful by OpenAI, Meta, Google, etc., into their frontier models. 

DeepSeek’s strategy of utilizing open-source models tin person a immense interaction connected the AI assemblage astatine large, opening up the AI marketplace and providing entree to AI tools for a wide acceptable of users, particularly smaller businesses. – Anuj Bhatia

2. MoE and MLA: DeepSeek-V3 stands retired owed to its architecture, known arsenic Mixture-of-Experts (MoE). In MoE models, aggregate specialized models collaborate to reply questions alternatively than relying connected a azygous ample exemplary to grip everything. Additionally, the exemplary employs a caller method called Multi-Head Latent Attention (MLA), which enhances ratio and reduces the costs of grooming and deployment. This enables DeepSeek-V3 to vie with immoderate of the astir precocious models disposable today.

3. Reinforcement Learning: DeepSeek’s occurrence tin beryllium attributed to a conception known arsenic reinforcement learning. This attack allows AI models to larn done proceedings and error, improving themselves done algorithms. It is rather akin to however humans larn from their experiences. Essentially, DeepSeek’s models larn by interacting with their situation and receiving feedback based connected their actions. As a result, these AI models go amended astatine reasoning and are susceptible of solving analyzable problems. Additionally, DeepSeek has excelled successful distilling the capabilities of its ample models into smaller, much businesslike ones.

4. Test-Time Compute: The DeepSeek-R1 exemplary is different offering from DeepSeek, featuring a unsocial capableness known as test-time compute, which allows it to ‘think’ portion generating responses. R1 utilizes the aforesaid Mixture-of-Experts (MoE) architecture and often matches oregon surpasses OpenAI’s apical exemplary successful areas specified arsenic mathematics, coding, and wide knowledge. Unlike OpenAI’s O1 model, which takes clip to process prompts and make optimal responses, R1 demonstrates its reasoning process successful existent time, revealing its concatenation of thought arsenic it produces output.

Deepseek app is seen successful this illustration taken connected Tuesday. (Photo: Reuters)

Hence, 1 tin spot that DeepSeek has fundamentally delivered a state-of-the-art exemplary that is competitive. Moreover, the institution has invited others to replicate their enactment by making it open-source. 

More cheap and affordable 

“OpenAI is highly overvalued. I deliberation we saw their concern exemplary benignant of stroke up implicit the past fewer days with DeepSeek fundamentally giving distant for escaped what they [OpenAI] wanted to complaint wealth for,” Gary Marcus, a prof astatine New York University (NYU), said successful an interrogation with CNBC on Tuesday, January 28.

DeepSeek’s AI models person been hailed arsenic a probe breakthrough arsenic they show that it is imaginable to make competitive, frontier AI models utilizing little currency and less GPUs – arsenic opposed to the billions of dollars spent by OpenAI, Meta, Google, Microsoft, and others to bash the same.

It is mostly believed that training AI models requires important investments. However, DeepSeek has minimised the tremendous costs associated with infrastructure and hardware. By utilising NVIDIA H800, which is considered an older procreation of GPUs successful the United States, DeepSeek has dramatically reduced the expenses related to gathering its AI models. In contrast, large American AI companies person opted for the much precocious NVIDIA H100 GPUs. DeepSeek, however, chose the little almighty version—NVIDIA H800, which is reported to person little chip-to-chip bandwidth.

Additionally, the DeepSeek-R1 exemplary is reported to beryllium 90-95% much affordable than OpenAI’s exemplary O1. Another important facet of gathering AI models is training, which requires important resources. According to the probe paper, the Chinese AI institution has lone trained the indispensable parts of its exemplary utilizing a method called Auxiliary-Loss-Free Load Balancing.

DeepSeek has undoubtedly shaken the satellite of AI, challenging the long-standing dominance of the US, which has led the AI contention with large players similar OpenAI and Google. Now, China’s DeepSeek is changing the landscape. But where does India basal successful this contention for AI dominance? More importantly, what tin India larn from the DeepSeek breakthrough to signifier its ain AI future?

Question 3: What are the lessons for India from the DeepSeek breakthrough?

DeepSeek’s AI models person not lone given Western AI giants a tally for their money but besides sparked fears that the US whitethorn conflict to support its AI primacy successful the look of a brewing tech acold warfare with China.- Karan Mahadik

DeepSeek’s technological achievement has stunned the world, from Silicon Valley to the planetary AI stage. However, China’s increasing dominance successful AI raises captious questions astir India’s position, particularly fixed the deficiency of an AI laboratory oregon startup that rivals the capabilities of OpenAI oregon DeepSeek. In this context, India tin gully cardinal lessons from the DeepSeek breakthrough:

1. AI is everyone’s game: DeepSeek’s breakthrough successful the AI tract demonstrates that if foundational AI models tin beryllium trained cost-effectively, it lowers the barriers for nations anxious to make their ain models. By reducing the fixed costs associated with gathering these models, resource-constrained countries tin amended compete, contempt challenges specified arsenic constricted GPU availability and insufficient backing for some foundational models and the required data. This means that AI is an accessible domain for everyone. Regardless of which state leads successful this area, immoderate federation tin reap the benefits of the AI race.

Sarjan Shah writes- As the Chinese breakthrough shows, necessity is so the parent of invention. By proving that advancement does not beryllium connected monolithic resources, this improvement offers anticipation that AI tin beryllium a instrumentality for everyone — not conscionable the fewer with billions to spend. It’s a reminder that successful the end, quality — whether artificial oregon quality — is astir reasoning differently.

2. Money buys galore things, but it is not everything/Wealth Builds, But Wisdom Innovates: While AI companies typically necessitate billions successful investments to bid their models, DeepSeek’s innovation showcases the effectual usage of constricted resources. This proves that groundbreaking advancements are not solely babelike connected backing but besides connected imaginativeness and adaptability. It besides reinforces however necessity tin thrust innovation successful unexpected ways. Thus, promoting probe and improvement is cardinal to resourceful innovation.

DeepSeek is simply a reminder that wealth buys galore things, but it is not everything; surely not emotion oregon the capableness for innovation. If the US is seen arsenic pouring billions of dollars into gathering much computing powerfulness and amended AI models, DeepSeek has shown it is imaginable to bash much with less.- C. Raja Mohan

C. Raja Mohan writes-  That should springiness anticipation for India and different mediate powers similar France. The spread betwixt the US and China is overmuch little than that betwixt the 2 of them and the rest.  While the mediate powers can’t support gait with the US and China, they could bash capable to enactment successful the AI  game.

3. AI Diplomacy-Navigating Partnerships for Progress: The Chinese breakthroug successful AI correspond not lone a technological breakthrough but besides a important geopolitical development. In this context, India should beryllium unfastened to collaboration with different countries to leverage the benefits of advancements successful AI.

C. Raja Mohan writes- The Biden Administration was unfastened to concern with India, but concerns astir leakage of exertion from India to Russia had enactment a dampener connected the benignant of entree the US is consenting to connection India connected AI chips. For Delhi, the clip is present to determine connected however overmuch value it is consenting to springiness Russia successful advancing its interests with the US connected the AI game.

Another important acquisition from the DeepSeek breakthrough is that the Chinese institution achieved this milestone contempt sanctions that constricted their entree to precocious chips and cutting-edge hardware. So, here’s a question—what is the antithetic country’s attack to regulating AI?

Question 4: How are countries regulating AI?

With the advancement of AI technology, governments and policymakers astir the satellite are progressively focused connected artificial intelligence. Nations look to beryllium successful a relentless contention to outpace each other, believing that those who autumn down volition yet beryllium the losers. However, galore acrophobic voices among them person valid reasons for their worries.

Geoffrey Hinton, a pioneer successful the tract of AI, has highlighted the imaginable for AI to surpass quality intelligence abilities. He has said that AI “will beryllium comparable to the concern revolution,” but helium has besides warned that “we besides person to interest astir a fig of imaginable atrocious consequences, peculiarly the menace of these things getting retired of control.” Thus, it is indispensable to retrieve however disruptive caller technologies tin beryllium for societies and economies.

The concerns regarding advancements successful AI tin beryllium categorized into 3 main areas: privacy, strategy bias, and violations of intelligence spot rights. Interestingly, argumentation responses to these issues alteration crossed antithetic jurisdictions.

The European Union has taken a notably stricter attack by proposing regulations that classify AI based connected circumstantial usage lawsuit scenarios, assessing them according to their level of invasiveness and risk. In contrast, the UK adopts a decidedly ‘light-touch’ approach, aiming to promote innovation alternatively than hinder it successful this emerging field.

The United States’ attack is positioned determination successful betwixt these 2 extremes, with indications of imaginable further deregulation. China has besides introduced its ain measures to modulate AI.

India has emphasized the request to code the challenges posed by the weaponization of societal media, advocating for steps to guarantee that AI promotes information and trust, adjacent arsenic the exertion presents important opportunities.

On the planetary stage, 28 countries, including the United States, the United Kingdom, and China, person agreed to the Bletchley Declaration. This marks the archetypal planetary statement aimed astatine addressing the risks associated with precocious artificial quality (AI). The declaration lays retired plans for greater transparency from AI developers regarding information practices and much technological collaboration connected knowing AI’s risks.

Prime Minister Narendra Modi is acceptable to co-chair the Paris AI Summit and has accepted the invitation to question to France. PM Modi with French President Emmanuel Macron. (File photo)

The regularisation of AI is continuously evolving successful effect to advancements successful the field. Each country’s attack to AI regularisation is thing we should intimately watch, particularly arsenic the AI scenery becomes much fascinating than ever. By the way, person you heard astir Kimi K1.5? What are your thoughts connected it? Let america cognize successful remark section. 

Post Read Questions

Prelims

(1) With the contiguous authorities of development, Artificial Intelligence tin efficaciously bash which of the following? (UPSC CSE 2020)

1. Bring down energy depletion successful concern units

2. Create meaningful abbreviated stories and songs

3. Disease diagnosis

4. Text-to-Speech Conversion

5. Wireless transmission of electrical energy

Select the close reply utilizing the codification fixed below: 

(a) 1, 2, 3 and 5 only

(b) 1, 3 and 4 only

(c) 2, 4 and 5 only

(d) 1, 2, 3, 4 and 5

(2) Consider the pursuing statements with respect to the Bletchley Park Declaration:

1. The declaration was signed by 28 countries.

2. Frontier AI is defined arsenic highly susceptible instauration generative AI models that could person unsafe capabilities that tin airs terrible risks to nationalist safety.

3. The United States, China, Japan, the United Kingdom, France, and India are not signatories to the declaration.

How galore of the statements fixed supra is/are correct?

(a) Only one

(b) Only two

(c) All three

(d) None

Mains

What are the main socio-economic implications arising retired of the improvement of IT industries successful large cities of India? (UPSC CSE 2021)

(Sources: Deepseek: How open-source AI is disrupting large tech’s monopoly , How DeepSeek’s origins explicate its AI exemplary overtaking US rivals ,DeepSeek’s emergence could marque OpenAI the WeWork of AI, Liang Wenfeng, Is this China’s ChatGPT infinitesimal and a wake-up telephone for the US?, What’s connected the docket of the Paris AI summit, DeepSeek’s Sputnik moment, In DeepSeek breakthrough, lessons for India)

For your queries and suggestions constitute at roshni.yadav@indianexpress.com

🚨New Year Special: Click Here to work the January 2025 issue of the UPSC Essentials monthly magazine. Share your views and suggestions successful the remark container oregon at manas.srivastava@indianexpress.com🚨

Subscribe to our UPSC newsletter and enactment updated with the quality cues from the past week.

Stay updated with the latest UPSC articles by joining our Telegram transmission – Indian Express UPSC Hub, and travel america on Instagram and X.