Mint Primer: AI power shift: Can China close the gap with the US?

Technology

Mint Primer: AI power shift: Can China close the gap with the US? – Crypto News

Published

14 hours ago

March 14, 2025

Dripp

If Chinese AI lab Deepseek shook the industry with its open-source, budget-friendly reasoning model earlier this year, another Chinese startup, Monica, is impressing many with its new general-purpose AI agentic model, Manus. Are there lessons for India?

What’s Manus and why is it the news?

Chinese artificial intelligence (AI) lab DeepSeek disrupted AI with its low-cost reasoning R1 model this year, prompting OpenAI to respond with o3 mini and GPT-4.5. Now, another Chinese startup Monica has introduced Manus, an autonomous AI agent that’s been dubbed the “second DeepSeek moment”. Manus can not only analyse and plan but also execute complex tasks without human intervention. In just seven days, its waiting list has grown to 2 million. Currently in beta, you need an invitation code to access Manus, a model that signals China’s growing challenge to the US’s AI dominance.

Also read | Can Musk factor, brand hype, Airtel-Jio partnerships boost Starlink’s services?

What’s special about AI agents?

‘Non-agentic’ systems like basic chatbots respond with scripted answers. AI agents initiate actions and take their own decisions. Consider an AI trading system that analyses market trends, makes investment decisions and adjusts its strategy based on performance. They have become smarter and more efficient with generative AI (GenAI) models that help them generate and respond to users in natural languages. Gartner expects AI agents to handle 80% of customer service issues by 2029, cutting operational costs by 30%. Anthropic CEO Dario Amodei predicts AI will generate nearly all codes within a year.

How are these models faring?

Altindex.com shows despite DeepSeek gaining traction, ChatGPT clocked 40.5 million downloads in January—50% more than DeepSeek and Google Gemini together. DeepSeek boasted 17.6 million downloads, followed by Gemini (9.6 million) and ByteDance’s Duobao (8.9 million). Interest in AI agents has surged in February, driven by the launch of Manus.

Also read | OpenAI discussing localization of ChatGPT India data

Do we need to temper expectations?

Manus is still in testing and crashes under heavy load. Its ‘Standard’ mode is fast, ‘Chain of Thought’ reasoning slower and power-intensive. AI agents struggle with multi-step reasoning and long-term memory, often making mistakes in complex tasks. Ethical and security risks, including phishing, misinformation and cyberattacks, raise concerns. Trust is another issue—flawed AI-driven financial or healthcare decisions could have serious consequences. This means governments must regulate to ensure accountability.

What can India do to keep up?

India is building its own graphics processing units (GPUs) and has built a datasets platform, AI Kosha, to help startups develop local language models while accessing GPUs at a lower cost. India also hopes to strengthen AI-driven research, innovation and skills development with its new AI Compute Portal and FutureSkills initiatives. Models like DeepSeek, Qwen (Alibaba) and Manus signal a shift to high-performance AI at low costs, offering India opportunities to develop cost-effective, localized AI solutions.

Also read | DeepSeek vs Stargate Project: How AI spending is evolving, in charts