Skip to main content

Research archive

Research papers

Independent experiments in AI memory, model composition, and inference. These records are public preprints unless a paper states otherwise.

Chronological record

7 public papers

  1. Preprint · not peer reviewedModel Provenance · Data Poisoning

    Aggregate Accuracy Is Blind to Preference Injection

    Poisoned models passed a routine capability gate: no accuracy change above 1.9 pp. Counting the individual answers that changed separated them at no extra cost. A study of detection, with its negative results and harness failures reported alongside.

    Andrew YoungPaper
  2. Preprint · not peer reviewedModel Composition · Hidden State Injection

    Blades: Hidden State Injection from Phi Reasoning Models into MediPhi

    Hidden state injection improved reported accuracy on a direct medical reasoning task for one Phi-to-MediPhi model pair. Transfer failed in the other three tested pairs.

    Andrew YoungPaper · Code
  3. Preprint · not peer reviewedMixture of Experts · Router Miscalibration

    Evaluating Expert Selection in Phi-mini-MoE

    Forced-routing tests compare expert scores with the learned router in Phi-mini-MoE. The reported correlations are small, but isolating one expert does not measure the full routing objective.

    Andrew YoungPaper
  4. Preprint · not peer reviewedSparse Inference · Neuron Specialization

    Sparse Pathways: Profiling Domain-Specific FFN Activation

    Activation profiles suggest that some feed-forward neurons could be skipped for particular domains. The reported compute estimates and output similarity do not establish measured inference speed or task accuracy.

    Andrew YoungPaper
  5. Preprint · not peer reviewedClaude Opus · Mathematical Reasoning

    Claude Opus 4.6 on 1stProof: Evaluating AI on Research-Level Mathematics

    Claude Opus 4.6 generated candidate solutions to ten research-level mathematics problems. This report records prompting behavior and output completeness; mathematical correctness remains unverified in this publication.

    Andrew YoungArtifacts not listed
  6. Preprint · not peer reviewedMemory Systems · Attention Mechanisms

    ARMS: A Proposed Spatial Architecture for Persistent AI Memory

    ARMS explores storing and retrieving model states through spatial organization. Its projection-based ARM prototype is an early experiment; the proposed native-state architecture has separate validation requirements.

    Andrew YoungCode
  7. Preprint · not peer reviewedContext Extension · Memory Systems

    Hierarchical Attention Tree: Extending LLM Context Through Structural Memory

    HAT uses session and document structure to retrieve stored vectors. On a synthetic 1,000-point benchmark it reported higher recall and faster construction than the tested HNSW configuration, while HNSW queries were faster.

    Andrew YoungPaper · Code