929: Dragon Hatchling: The Missing Link Between Transformers and the Brain, with Adrian Kosowski
Breaking news: Jon Krohn welcomes Adrian Kosowski to the show to talk about the groundbreaking research happening at Pathway. Adrian and his team demonstrate how they have brought attention in AI closer to the way the brain functions, creating, in essence, a “massively parallel system of [artificial] neurons” that communicate with one another and exhibit properties similar to natural neurons. The goal is to move beyond the current limitations of transformers, where reasoning can be generalized across more complex and extended reasoning patterns, approximating a more human-like approach to problem-solving.
This episode is brought to you by the Trainium2, the latest AI chip from AWS, by Dell, by Intel, by and Gurobi.
Additional materials: www.superdatascience.com/929
Interested in sponsoring a SuperDataScience Podcast episode? Email natalie@superdatascience.com for sponsorship information.
In this episode you will learn:
- (01:27) Pathway’s ground-breaking new biologically inspired architecture
- (20:40) Limitless context windows
- (34:39) BDH architecture as positive space
- (53:11) Building multilingual models
- (1:01:07) How to access the BDH architecture
