BriLLM-MetaPred
The first language model derived from a mathematical theory of AGI. It is based on the MetaPred primitive and the SiFu learning paradigm and employs a single dynamical state that serves simultaneously as the context representation and the source for predictive readout, with training and inference symmetrically coupled.
Timeline (last 90 days)
BriLLM-MetaPred is the first engineering implementation of the MetaPred machine learning framework.
BriLLM-MetaPred is the first machine learning paradigm derived entirely from mathematical theory, rather than assembled through manual design and incremental improvement.
BriLLM-MetaPred uses a single, continuously evolving dynamical state that simultaneously serves as the contextual representation and the state source for predictive readout.
BriLLM-MetaPred continuously measures its own dynamical state to reduce the inconsistency caused by finite depth, maintain internal consistency, and provide an endogenous dynamical self-supervised signal in the form of a fixed point.
BriLLM-MetaPred is neither a recurrent language model nor a simple extension of BriLLM, but a bridge connecting the MetaPred primitive and BriLLM's SiFu learning paradigm.
BriLLM-MetaPred's predictive performance is comparable to GPT baselines of the same scale.
BriLLM-MetaPred is the first machine learning paradigm fully derived from mathematical theory, and it reaches the level of same-scale GPT on language modeling tasks.
All Facts (7)
BriLLM-MetaPred is the first engineering implementation of the MetaPred machine learning framework.
50%UnverifiedBriLLM-MetaPred is the first machine learning paradigm derived entirely from mathematical theory, rather than assembled through manual design and incremental improvement.
50%UnverifiedBriLLM-MetaPred uses a single, continuously evolving dynamical state that simultaneously serves as the contextual representation and the state source for predictive readout.
50%UnverifiedBriLLM-MetaPred continuously measures its own dynamical state to reduce the inconsistency caused by finite depth, maintain internal consistency, and provide an endogenous dynamical self-supervised signal in the form of a fixed point.
50%UnverifiedBriLLM-MetaPred is neither a recurrent language model nor a simple extension of BriLLM, but a bridge connecting the MetaPred primitive and BriLLM's SiFu learning paradigm.
50%UnverifiedBriLLM-MetaPred's predictive performance is comparable to GPT baselines of the same scale.
50%UnverifiedBriLLM-MetaPred is the first machine learning paradigm fully derived from mathematical theory, and it reaches the level of same-scale GPT on language modeling tasks.
50%