7 papers
How Transformers Reject Wrong Answers: Rotational Dynamics of Factual Constraint Processing
Javier MarÃn
When a decoder-only transformer is forced to process matched correct and incorrect single-token continuations of a factual query, the two pathways through hidden-state space diverg…
A Geometric Taxonomy of Hallucinations in LLMs
Javier MarÃn
Hallucinations in deployed language models can have real consequences for downstream decisions in domains such as healthcare, legal, and financial services. In production, detectio…
Semantic Grounding Index: Geometric Bounds on Context Engagement in RAG Systems
Javier MarÃn
When retrieval-augmented generation (RAG) systems hallucinate, what geometric trace does this leave in embedding space? We introduce the Semantic Grounding Index (SGI), defined as…
Empirical Characterization of Temporal Constraint Processing in LLMs
Javier MarÃn
When deploying LLMs in agentic architectures requiring real-time decisions under temporal constraints, we assume they reliably determine whether action windows remain open or have…
Loss Given Default Prediction Under Measurement-Induced Mixture Distributions: An Information-Theoretic Approach
Javier MarÃn
Loss Given Default (LGD) modeling faces a fundamental data quality constraint: 90% of available training data consists of proxy estimates based on pre-distress balance sheets rathe…
Capability Ceilings in Autoregressive Language Models: Empirical Evidence from Knowledge-Intensive Tasks
Javier MarÃn
We document empirical capability ceilings in decoder-only autoregressive language models across knowledge-intensive tasks. Systematic evaluation of OPT and Pythia model families (7…