I believe that when looking for ideas—such as developing an agent to find new directions—the content searched and learned should not be computer-related. At the minimum, 'agent' should not be the keyword. Political science and management are viable directions for agent teams.
For instance, self-evolution should not be about self-repair. Instead, it should use a bottom-tier elimination process to remove specific agents, on the condition that each agent has different prompts and organizational forms. From a scientific perspective, this is Top-K sampling
Peek‘s prediction is always right.
1. Enhancing human and replacing human are different logic, replacing human is risk control, now harness engineer are popurlar.
2. Manus's team consists of three groups. sandbox, eval, architecture
Introducing Lavis — your AI assistant that can see and operate your Mac for you.
Tell it what you want, and it will look at your screen, find the right buttons and fields, click, type, and get things done — just like a person sitting in front of your computer.