On a North Atlantic Ocean night of April 14, 1912, RMS Titanic glided through the calm waters. Iceberg warnings from SS Californian and Mesaba didn’t reach the bridge promptly as the wireless operators prioritized passenger communications. Capt. Edward Smith kept the ship going at a high speed to maintain a swift crossing despite awareness of general ice conditions.
Deeper into the night, lookout Frederick Fleet spotted an iceberg directly ahead and sent an urgent warning. First officer Willian Murdoch ordered the ship to turn and engines to reverse. However, the ship’s starboard side stuck the iceberg causing damage to the hull owing to the massive size of the ship and the momentum it had. Titanic sank within hours taking with it over 1500 passengers and crew. Delayed communication, human error, and a slow response to known risks culminated in a tragedy of colossal proportions, prompting transformative changes in maritime safety regulations—like the implementation of continuous radio watch.
Reminiscent of the iceberg warnings, Venturebeat ran an article On Sep 10, 2024, titled “DeepSeek-V2.5 wins praise as the new, true open source AI model leader” that waxed eloquent about DeepSeek-V2, a high-performance, cost-efficient LLM garnering praise for being an accessible alternative to proprietary models, offering competitive performance at a lower cost.
The piece highlighted DeepSeek-V2 as a catalyst for change, shifting the balance in AI development toward more open, cost effective and accessible solutions, in contrast to the dominance of proprietary models from large tech companies. While the DeepSeek-V2.5 challenged the existing Chinese AI incumbents such as ByteDance, Tencent, Baidu and Alibaba to lower their AI model prices within China, it failed to get the appropriate attention from the world press and the industry.
The Titanic moment arrived four months later on January 20, 2025, when DeepSeek released their open-source reasoning model, DeepSeek-R1, claimed to be at par with OpenAI’s most advanced LLM, o1 with purported training costs well under $6 million.
DeepSeek’s innovative approach involves using lower-cost hardware (2,000 Nvidia’s H800 GPUs) to train a high-performance AI model at a fraction of the cost of current industry leaders. AI observers might have been blindsided by a bouquet of other LLM releases around the world as the U.S. markets fell by a big scoop ($1 trillion wiped off Nasdaq) reminding one of the Titanic sinking.
Amidst the din of the Big Tech “Stargate” and LLM announcements, the Tech media missed the impressive portfolio of DeepSeek LLMs and image generation models that had preceded DeepSeek-R1. Alibaba cloud followed soon with the release of their Qwen 2.5 – Max model with competitive results against industry leaders like GPT-4o and Claude-3.5-Sonnet in tests of advanced reasoning and knowledge.
These models are one of many LLMs coming out of China as it doubled its LLMs from 79 in 2024 to 200 by the end of 2024 executing its 2017- A New Generation Artificial Intelligence Development Planthat details a plan to make China the world’s major AI innovation centre by 2030 with open sourcing and tighter industry academia collaboration as key tenets of the policy.
This intense work has translated into a clear leadership position on the Generative AI patents, with China leading U.S. and other countries by a huge margin across all type of models with Large Language Models (LLMs) seeing the highest growth (53 in 2020 to 881 in 2023).
Learnings for India
India’s AI Plan differs from China in its focus on Social and inclusive growth (NEP 2020 has a passing reference to AI) focus in contrast to China’s competitive AI leadership plan. The Chinese Artificial Intelligence Innovative Action Plan for Institutions of Higher Education (2018) placed colleges and universities as the main force behind building the world’s main AI innovation centres and in leading the development of a new generation AI talent pool to provide China with the scientific and technological support and guaranteed talent to put it at the forefront of innovation-oriented countries.
An ORF critique aptly points towards the inward-oriented NEP that prioritises “institutional restructuring and consolidation” and a “more holistic education” that is mindful of multi-faceted human capacities. While India has done well with laying down the core blocks of the India AI mission with its “Few, Many and All” approach towards AI skilling along with compute and data infrastructure, it can borrow a leaf from the China playbook on the following parameters:
Clear targets
India AI mission can promulgate clear targets for the Higher Education Institutions such as those around AI patenting (Chinese academia occupy the top 9 Gen AI patent owner spots) and curriculum development. This can be development of 100 “AI + X” majors for interdisciplinary growth, 50 world-class AI textbooks and opening of 50 national-level AI courses along with annual training of 500 teachers and 5,000 students in AI by 2020.
Deepen focus on creativity and critical thinking
In contrast to his competitors, Liang, the 40-year-old DeepSeek founder has stressed the importance of basic skills, creativity and passion as opposed to mere degrees or certifications. In a 2023 interview with a Chinese media outlet 36Kr, he emphasized the importance of hiring Gen Z and humanities majors to augment the lateral thinking skills in his team. He said the startup consisted of “mostly fresh graduates from top universities, PhD candidates in their fourth or fifth year, and some young people who graduated just a few years ago.”
The Indian Centres of Excellence in AI for Education, announced in the recent budget can enhance personalized learning, curriculum design, and pedagogy and create competency-based assessments that test higher-order thinking skills such as analysis, critical thinking, and problem-solving.
Increase open-source experimental research in collaboration with industry
The DeepSeek saga has taught the world the value of experimentation with alternate techniques with resource constraints. India’s performance on experimental research is way below China and US ( 19.5% of overall R&D expenditure Vs 82% in China and 65% in USA). India AI Mission’s call for proposals can increase the pace of experimentation preferably in an open-source philosophy, thereby lowering the barriers to entry, allowing smaller companies and independent developers to compete without requiring billion-dollar budgets. ANRF and other institutional frameworks should be leveraged to provide grants for open ended research to allow commercialization and creation of independent research labs such as AI4Bharat.
Create full stack of AI engineering capabilities
The future AI engineers require hardware and assembly level programming capabilities in addition to the software stack. The DeepSeek engineers, operating at an exceptional level, showcased x100 proficiency in both complex hardware and low-level programming. By bypassing CUDA, they optimized NVIDIA chips at an even deeper layer thereby working around the U.S. chip restrictions that affect India too. India urgently needs to build an army of full stack AI engineers, who have integrated AI hardware and software capabilities to address hardware restrictions borne out of geopolitical considerations.
Retain top AI talent within country
According to the Global AI Talent Tracker 2.0 by Marco Polo, over the past few years, China and India have significantly expanded their domestic AI talent pool to support the burgeoning AI industry. Although India has done well to retain 20% of those researchers in India, a good lot (60%) of them still leave to the U.S. for graduate school and work there later as they are drawn to the premier research institutions and sophisticated lab infrastructure
Much like the U.S., it’s important to provide the smartest AI researchers in the country with the best resources to conduct open ended research with minimal bureaucracy. Top talent tends to congregate around other top talent, so this model can lead to a positive self-affirming loop. India needs to amplify its ability to attract AI talent with a slew of targeted measures to increase its talent base beyond a few hundreds to tens of thousands of AI researchers.
India, with its strong talent and technology base, must proactively invest in AI pedagogy, research, infrastructure, and policy framework to emerge as a formidable force in the global AI arena by 2030.
(The author is an Emerging Technology expert with experience in setting up DeepTech public private partnerships and policy advisory in areas of AI, IoT, 5G, Geospatial, Autonomous and Data Centre Technologies)
Published – February 06, 2025 09:54 pm IST
