Joscha Bach Λ Ben Goertzel: Conscious Ai, LLMs, AGI
Theories of Everything with Curt Jaimungal
Expanding Dimensions and Limitless Capabilities of AI Training Data
The training data for language models like LLM can have dimensions in the range of few hundred thousand, making it challenging to add new dimensions or significantly extend the existing ones. However, new combinations within the dimensions are possible. Language models are not limited to the discovered dimensions and can be configured to explore and verify new dimensions. They have the ability to use plugins to write code, use compilers, cameras, sensors, and actuators to conduct experiments. The transformer algorithm focuses on finding the most likely token, which can result in suboptimal decisions when it makes mistakes. This approach contrasts with the idea of finding the best possible token. Language models can be prefaced as simulations aiming to approximate the search for the best token in a mind.
Remember Everything You Learn from Podcasts
