Recently, the DeepMind team led by ShaneLegg has released a paper on AGI, proposing six criteria and five levels for AGI, and pointing out that modern large language models are taking initial steps towards the embryonic form of AGI. The team predicts a 50% chance of developing AGI by 2028. This paper provides a clear roadmap and timeline for the development of AGI, offering strong support for research and discussions in this field.