With the advent of powerful large language models, or LLMs, many are asking if real AGI (artificial general intelligence) is on the horizon.
In order to create AGI, we need equations that are more powerful – and more comprehensive than those being used now in LLMs.
Current LLMs use encoder/decoder algorithms, which are based on transformer equations.
The equations that will truly support AGI will come from active inference, an area developed by Friston and colleagues.
See this Themesis blogpost (plus the dozen leading up to it) for a resource compendium on the Kullback-Leibler equation, free energy, and active inference.
Themesis will be offering its own unique equations – ones that will contribute to a new neural networks class, and potentially support AGI – in the near future.
Alianna J. Maren, Ph.D.
Founder and Chief Scientist, Themesis, Inc.