The breakthrough performance of the Manus model prompts reflection on the development path of AI

Recently, the Manus model achieved breakthrough results in the GAIA benchmark test, surpassing other large language models of the same level in performance. This means that Manus can independently handle complex tasks such as multinational business negotiations, involving multiple aspects like contract clause analysis, strategy formulation, and proposal generation, and even coordinate legal and financial teams.

The advantages of Manus are mainly reflected in three aspects: dynamic goal decomposition ability, cross-modal reasoning ability, and memory-enhanced learning ability. It can decompose complex tasks into hundreds of executable subtasks, handle various types of data simultaneously, and continuously improve decision-making efficiency and reduce error rates through reinforcement learning.

This breakthrough has once again sparked discussions in the field of artificial intelligence about the future development path: should we move towards the direction of Artificial General Intelligence (AGI), or should multi-agent systems (MAS) take the lead in collaboration?

The design philosophy of Manus suggests two possibilities: one is the AGI path, which involves continuously enhancing the intelligence level of individual units to approach human comprehensive decision-making capabilities; the other is the MAS path, which acts as a super coordinator directing thousands of specialized intelligent agents to work collaboratively.

On the surface, this is a debate about technological pathways, but it actually reflects a deep contradiction in how efficiency and safety are balanced in the development of AI. The closer a single intelligence gets to AGI, the higher the risk of opacity in its decision-making process; while multi-agent collaboration can disperse risks, it may miss critical decision-making opportunities due to communication delays.

The progress of Manus has also magnified the inherent risks in AI development, including issues such as data privacy, algorithmic bias, and adversarial attacks. For example, in medical scenarios, Manus needs access to sensitive patient data; in financial negotiations, there may be undisclosed corporate financial information involved. In the recruitment process, there could be salary discrimination against specific groups; during the legal contract review, there might be a higher error rate in judging the terms of emerging industries. Additionally, hackers may interfere with Manus' judgment in negotiations by embedding specific audio frequencies.

These challenges highlight a concerning fact: the more intelligent AI systems become, the broader their potential attack surface.

In the Web3 space, security has always been a core focus. Under the framework of the "impossible triangle" proposed by Ethereum founder Vitalik Buterin (the idea that a blockchain network cannot simultaneously achieve security, decentralization, and scalability), various cryptographic technologies have emerged:

1. Zero Trust Security Model: Based on the principle of "trust no one, always verify," it rigorously authenticates and authorizes each access request.

2. Decentralized Identity (DID): A new type of decentralized digital identity standard that enables identity verification without a central registration authority.

3. Fully Homomorphic Encryption (FHE): Allows computation on encrypted data, protecting data privacy while realizing data value.

Among these technologies, FHE is considered an important tool for addressing security issues in the AI era. It can provide protection on multiple levels:

• Data Layer: All information input by users (including biometric features, voice, etc.) is processed in an encrypted state, meaning that even the AI system itself cannot decrypt the original data.

• Algorithm level: Achieving "encrypted model training" through FHE to ensure that the AI decision-making process is not exposed to external scrutiny.

• Collaborative level: Communication between multiple agents uses threshold encryption, so the compromise of a single node does not lead to global data leakage.

Although Web3 security technologies may not have a direct connection to ordinary users, they have a profound impact on user interests. In this challenging environment, it is essential to continuously improve security measures.

Some projects have made progress in these areas. For example, uPort launched a decentralized identity solution in 2017, and NKN released a mainnet based on a zero-trust model in 2019. In the field of FHE, Mind Network became the first project to go live on the mainnet and has partnered with several well-known institutions.

As AI technology approaches human intelligence levels, non-traditional defense systems are becoming increasingly important. Technologies such as FHE not only address current issues but also lay the foundation for the future era of strong AI. On the path to AGI, these security technologies have become indispensable elements.