My work focuses on expanding the one-directional inference nature of LLMs to allow for a more nuanced view of complex concepts, such as memories that evoke both sorrow and joy. For example, I have found that LLMs often overfit to singular emotional labels when prompted with personal memories, failing to capture their multifaceted impact. This limitation prompted me to experiment with approaches inspired by physics.
The ontological inversion, a few months back when I was obsessed with turning physics into a language and I was able to invert a concept almost as an inversion or involution. I didn’t realize what I stumbled upon at the time it came really just from apply a “negative gain factor” during one of my sweeps into Niodoo steering architecture. Since I don’t come from traditional AI background apologies upfront if this is a thing but if it helps solves someone’s problem or pushes llms to understanding better i thought it better to be shared than collecting dust in my drive. Excuse the terms i use also i know the immediate slop goggles are on. I don’t use traditional phrases because to be honest I don’t know them, and I don’t go searching for them. I hope you guys can look past that.