Keep pulling the thread on Yann LeCun & Alex LeBrun.
Future AI assistants will be always-on companions, likely embedded in smart glasses, that will observe a user's environment to provide continuous advice and enhance their intelligence.
Advanced Machine Intelligence Labs (AMI Labs), founded by Yann LeCun, plans to build "world models" as a successor to large language models.
To unlock new AI use cases, a different approach from LLMs is needed, as LLMs are a 'shortcut' focused only on language and represent a small part of the brain's functionality.
Developing off-the-shelf, universal world models is expected to take three to five years, or possibly more.
AI-accelerated MRI technology, which originated from a collaboration between NYU and Meta's FAIR lab, is now in basically every MRI machine on the market.
AI assistants will amplify human capabilities rather than replace them, with humans remaining in control.
Current AI agent systems are brittle and only work in specific cases because they are trained like LLMs to imitate human action sequences from vast amounts of data.
The limitations of Large Language Models are particularly acute in high-reliability domains like healthcare.
The total amount of publicly available text on the internet is about 30 trillion words, which translates to approximately 10^14 bytes of data.
A four-year-old child processes approximately 10^14 bytes of visual data, an amount equivalent to all the text on the public internet.
Generative AI approaches that work well for language do not work for video or other sensory inputs because the real world is high-dimensional, continuous, and complex.
Advanced Machine Intelligence Labs (AMI Labs) expects to achieve results on narrow, specific use cases for world models within one to two years by working with industrial partners.