Model Accuracy Challenges
Models assign probability distributions over sequences, which can lead to errors even with high accuracy. As the conversation explores the balance between training and inference compute, the potential for models to achieve human-like performance becomes evident, emphasizing the need for logical reasoning and redundancy in their design. The hope is that with increased computational power, the probability of mistakes will significantly diminish.In this clip
From this podcast
No Priors
No Priors Ep. 74 | With Google DeepMind VP of Research Oriol Vinyals
Related Questions
Will large language models scale all the way to artificial general intelligence (AGI) as discussed in the episode Oriol Vinyals: Deep Learning and Artificial General Intelligence | Lex Fridman Podcast #306 and the clip AGI and Society?
Will large language models scale all the way to artificial general intelligence (AGI) as discussed in the episode Oriol Vinyals: Deep Learning and Artificial General Intelligence | Lex Fridman Podcast #306 and the clip Scaling AI Methods?
What is the role of large models in predicting the next word in language tasks as discussed in the episode Ilya Sutskever: Deep Learning | Lex Fridman Podcast #94 and the clip History of Neural Networks in Language?