Learn about the challenges of filtering malicious signals in self-supervised learning, and how Facebook's Meta AI addressed this using only 17,000 tags corresponding to physical things or situations for image recognition systems.
The predictions of AI capabilities have been completely inverted and wrong, with AI now being able to generate art and talk about emotions. While the future of hybrid human-AI systems is interesting, the possible impacts on individuals born inside of it remain unknown.
The Transformer architecture has been widely used in neural networks, despite not being thoroughly explained in its original paper. Researchers speculate the design decisions and consider it a general-purpose computer.
Deep fakes, videos that use AI to make it look like someone is doing or saying something they never did, are becoming increasingly sophisticated and harder to detect.
The development of AI technology that is similar to humans requires a deeper understanding of human beings and their unique capabilities, including the ability to craft stories that take unexpected turns, according to the podcast discussion of The AI Alignment Podcast.