In less than a week, Manus AI, China’s new artificial intelligence agent, has captured the attention of more than 2 million users who have signed up for its waiting list. This enthusiasm has led some to call it the “second DeepSeek moment.” However, the reality behind this phenomenon is more complex than it seems.
Despite the euphoria, many analysts believe that the hype surrounding Manus AI is somewhat exaggerated. Although it is presented as China’s answer to OpenAI’s advancements, it is essential to look beyond the grandiose claims of certain social media influencers. Manus AI is a promising project, but not a true revolutionary breakthrough.
Why Manus AI is not a significant breakthrough
The key to DeepSeek’s success lay in its ability to replicate OpenAI’s reinforcement learning method, achieving impressive results on a tight budget. This team not only managed to innovate but also opened its training method, allowing other labs to train cutting-edge reasoning models. On the other hand, Manus AI relies on existing models, such as the Claude 3.5 Sonnet from Anthropic and several fine-tuned Qwen models.
While the integration and tools that Manus offers are undoubtedly advantages, the real breakthrough lies in creating models that can effectively perform complex tasks.
Currently, the Manus team is internally testing the unified model Claude 3.7 Sonnet, which seems promising. However, the real challenge remains the development of highly capable artificial intelligence models.
The stumbles of Manus AI
Despite the attention it has generated, some users who had early access to Manus AI have shared experiences that leave much to be desired. For example, Derya Unutmaz, a biomedical scientist, compared Manus with OpenAI’s Deep Research agent and found that while the latter completed a task in just 15 minutes, Manus took 50 minutes and failed to finish it. Moreover, Manus does not cite sources, which limits its usefulness compared to Deep Research.
Another user, teortaxesTex, pointed out that Manus is more efficient at repeating information than at performing agent tasks. In fact, a viral video showing Manus AI automating 50 tasks turned out to be false. Yichao ‘Peak’ Ji, the chief scientist of Manus, confirmed that the video did not represent the actual functioning of the agent.
Despite these drawbacks, it is important to remember that Manus AI is still in closed beta phase, so it is premature to dismiss it completely. However, it is essential to be cautious when trying new artificial intelligence products. Manus may not be a revolutionary breakthrough, but it is an ambitious start towards a future where artificial intelligence plays a more independent role in everyday tasks.
0 Comments