Summary

  • China-based startup Butterfly Effect has launched its general AI agent, Manus, to great interest and hype from some quarters, although few people have access to it.
  • Invoking comparisons to the AI model DeepSeek, Manus is designed to operate on a wide array of tasks rather than just conversations, and its “Manus’s Computer” window shows users what it is doing.
  • MIT Technology Review tested the AI with three tasks, including finding journalists covering China tech, searching for a New York two-bedroom property under $900k, and nominating innovators for MIT’s Under 35 list.
  • On the first task, Manus provided a short list of five journalists, which was expanded to 30 when pressed for consistency.
  • For the property search, Manus adapted well to clarified criteria after initially excluding some results.
  • For the third task, Manus performed poorly, providing three candidates for 50 required by the broad criteria and 30 required by the specified criteria.
  • Manus is intuitive and suitable for many users, but its instability, high failure rate, and propensity for crashes mean it is not yet widely applicable.

By Caiwei Chen

Original Article