Manus AI Does Not Represent China’s Next DeepSeek Milestone; Look Past the Hype
The Rise of Manus AI: A Closer Look
In just one week, over 2 million individuals have signed up for the waiting list to access Manus AI, a general AI agent developed in China. This surge in interest has generated talk of Manus AI being a significant event comparable to OpenAI’s Deep Research agent. However, at this stage, Manus is still in a closed beta phase, and invitations are needed to join.
Manus AI: Hype vs. Reality
While many people are excited about Manus AI, referring to it as a "breakthrough," some are skeptical about the extent of its innovation. Critics suggest that the excitement may be inflated, especially due to the influence of AI advocates on various social media platforms. Although Manus AI shows promise, it might not be the revolutionary step some claim it to be.
Comparison with DeepSeek
To understand the nuances of Manus AI, it’s important to compare it with DeepSeek. DeepSeek is noted for successfully emulating OpenAI’s reinforcement learning techniques while maintaining a lower budget for its operations. The development team at DeepSeek even released an open-source training method called GRPO, which allows other research teams to develop advanced reasoning models.
In contrast, Manus AI has integrated several existing technologies, including Anthropic’s Claude 3.5 Sonnet model and several fine-tuned Qwen models, utilizing the Browser Use open-source project. While Manus AI benefits from better integration and tooling, it does not represent the same kind of groundbreaking advancement that DeepSeek achieved.
Performance Insights: Early User Experiences
Even though Manus AI is still in its beta testing phase, some early users have shared their impressions online. For instance, Derya Unutmaz, a biomedical scientist, tested Manus alongside OpenAI’s Deep Research agent. He discovered that while Deep Research completed its assigned task in 15 minutes, Manus AI took 50 minutes but failed to finish. Notably, he pointed out that Manus AI does not reference sources, a feature that Deep Research provides.
Other users have also voiced their concerns regarding Manus AI’s capabilities. One user stated that Manus is better at repeating information than handling agentic tasks. Another user discovered that Manus failed to mention key topics, such as a major gaming console, during a research task.
There was also a viral video that claimed to showcase the Manus AI agent automating 50 tasks. However, Yichao ‘Peak’ Ji, the chief scientist behind Manus, confirmed that this video did not depict the actual Manus AI, dismissing it with a lighthearted comment.
Future Considerations for Manus AI
While Manus AI has faced initial setbacks, it’s crucial to remember that it is in a closed beta phase, making premature judgments inappropriate. The Manus team has already indicated that they plan to enhance the agent significantly before offering it to the general public. As AI technology continues to evolve, so too will the capabilities of products built on these advancements.
Manus AI may not yet be the game-changer that some anticipated, but it has opened a door to an exciting future of AI development. Observers will be keen to see if it can evolve in response to the challenges it’s currently facing and how it might refine its abilities over time.
