Chapter
Building AGI with neural modules
2:11:10 - 2:22:48 (11:37)
The way to build an AGI would be by taking 25 different neural network modules that resemble different parts of the brain and post-processing their output to make it more structured, according to Lex Fridman, an AI researcher and podcast host.
Clips
The transformer model can be used to train neural models on the robot Philip K. Dick, generating realistic senses including those that disobey the rules of grammar and producing pronouncements similar to those that Philip K. Dick might have said.
2:11:10 - 2:15:07 (03:56)
Summary
The transformer model can be used to train neural models on the robot Philip K. Dick, generating realistic senses including those that disobey the rules of grammar and producing pronouncements similar to those that Philip K. Dick might have said.
ChapterBuilding AGI with neural modules
Episode#103 – Ben Goertzel: Artificial General Intelligence
PodcastLex Fridman Podcast
The podcast discusses two approaches to learning implicit representations in neural networks, including treating them as an oracle and using GAN type neural networks to have a Bayes net of semantic latent variables with dependencies between them.
2:15:07 - 2:19:01 (03:54)
Summary
The podcast discusses two approaches to learning implicit representations in neural networks, including treating them as an oracle and using GAN type neural networks to have a Bayes net of semantic latent variables with dependencies between them.
ChapterBuilding AGI with neural modules
Episode#103 – Ben Goertzel: Artificial General Intelligence
PodcastLex Fridman Podcast
A community looking to build an AGI would use around 25 different neural modules that mimic different parts of the brain.
2:19:01 - 2:22:48 (03:46)
Summary
A community looking to build an AGI would use around 25 different neural modules that mimic different parts of the brain. Companies like DeepMind are more likely to build AGI due to their hiring of individuals with diverse AGI approaches while OpenAI is seen as a more pure deep reinforcement learning shot.