Yichao ‘Peak’ Ji, a 33-year-old Chinese entrepreneur and tech enthusiast known for creating the mobile browser Mammoth and founding Peak Labs in the US, is the cofounder and chief scientist of Manus AI.
While DeepSeek signalled China’s progress in large language models, Manus is a leap into AI autonomy, capable of executing multi-step workflows, accessing authoritative data sources through application programming interfaces (API).
Yichao Ji calls Manus “the first general AI agent” with the potential to solve real world problems. “We see it as the next paradigm of human-machine collaboration and potentially a glimpse into artificial general intelligence (AGI),” he said.
Autonomous AI agent
Backed by Chinese investors and developer Monica.im, Manus is currently available as an invitation-only web preview. From screening resumes, to conducting research or correlation analysis between stocks, it has varied use cases.
Discover the stories of your interest
- It leverages a multi-agent architecture, functioning like an executive managing a team of specialised sub-agents to break down and execute complex tasks efficiently.
- Its cloud-based asynchronous operation eliminates the need for constant user input, running seamlessly in the background and notifying users only when results are ready.
Towards AI autonomyGAIA serves as a benchmark for assessing General AI Assistants in tackling real-world challenges. Manus has achieved new state-of-the-art (SOTA) performance across all its three difficulty levels. The AI was evaluated using the same configuration as its production version for reproducibility.
“This isn’t just another chatbot or workflow. It’s a truly autonomous agent that bridges that gap between conception and execution. While other AIs stop at generating ideas, Manus delivers results,” Yichao Ji said while presenting the demonstration.
For instance, when given a prompt like “find me an apartment in San Francisco,” Manus analyses crime rates, rental trends, and user preferences to create an optimised shortlist without needing further input, as per use cases cited by the company.
The second DeepSeek moment
The launch of DeepSeek was seen as a “Sputnik moment” due to China’s ability to create efficient large language models (LLMs) at a fraction of the cost of its American counterparts.
Industry leaders in Silicon Valley quietly recognise that China’s aggressive advancements in autonomous systems could grant it a first-mover advantage in key sectors.
Manus AI further proves that such progress can be replicated. It follows the battle between OpenAI and DeepSeek, the most cost-efficient reasoning model.
The open-source availability of Chinese models allows more industries, including governments, healthcare, banking, and insurance, to adopt AI as they now have an on-premise alternative.
Implications for India
During the second day of his Microsoft India AI Tour in January, CEO Satya Nadella said that India can do great work in the area of Indic languages and transforming its industries using artificial intelligence.
As many as 91% of India-based companies will use half or more of their data to train AI models in 2024, ET reported. While training neural networks on extensive natural-language data requires significant computational power and engineering expertise, relying solely on APIs to connect clients with existing products is a cautious approach, as per a Bloomberg report.
Now is the moment to adopt a more ambitious approach, say experts—one that moves beyond merely customizing OpenAI’s GPT-4, China’s DeepSeek, Google’s Bard, or Meta’s LLaMA with specialised data for a fee.
India’s pragmatic approach
India will emerge as one of the key players not only in providing services but also as a major technology creator in the age of artificial intelligence, Union minister for information and broadcasting Ashwini Vaishnaw said at the 25th ET Awards for Corporate Excellence.
“We should have three foundational models by the end of this year. Developing our own GPU… we should be having in the next three to four years. Just a couple of days ago, we launched the 14,000 GPU common compute facility. That’s a common compute facility available for all startups, academia, entrepreneurs, researchers, everybody, Vaishnaw said.
ET reported earlier this month that the government is currently evaluating 67 applications for an Indian foundational model, of which 22 are large language models.
While challenges remain, noting the use cases for AI, India has a unique opportunity to leverage AI for social entrepreneurship in areas like agriculture and healthcare to solve key challenges of the developing world.
