Bookmark this to keep an eye on my project updates!
Good primer of reinforcement learning to imagination.
p
Real intelligence is about understanding the world through a sequence of patterns. We experience the world, store these experiences in our memory, and use them to predict what will happen next. This ability to constantly predict the future based on past experiences is what makes us intelligent.
p
Scott Phoenix proposes that AGI should also be able to imagine and mentally simulate the world. They point out that most of the connections in the brain go from higher-level areas to the visual cortex, which suggests that the brain is constantly predicting what it will see. This ability to imagine allows us to be surprised by unexpected events and to constantly update our models of the world.
Overall, the video argues that LLMs are a powerful tool, but they are not on the path to achieving AGI. We should not be fooled by their fluency into thinking they are intelligent.
The state of the art of ML undoubtedly works and is beneficial. But is it truly understandable?
p