Transformers in AI
Geoffrey and Craig delve into the innovative use of transformers for image recognition, highlighting how they facilitate interactions between image patches similar to word fragments in NLP models. They discuss the differences between supervised and unsupervised training in this context, emphasizing the transformative potential of capsule networks in generating new representations. The conversation also clarifies misconceptions about how models like GPT-3 utilize vast amounts of data, revealing that they synthesize information rather than merely matching it.In this clip
From this podcast
Eye on AI
Geoffrey Everest Hinton reviews his work in 2020 and talks about what he sees on the AI horizon
Related Questions