large language models

A local non-containment in Step (v) of IUT IV, Theorem 1.10

Step (v) of the proof of Theorem 1.10 in Mochizuki's Inter-universal Teichmüller theory IV bounds, for each tuple of places, the log-volume of the corresponding component of the holomorphic hull of the union of possible images of a Θ-pilot object defined in Corollary 3.12 of the third paper. This note gives an explicit...

Age Is Not Evidence of Staleness: A Recency Decay Is a One-Rate Survival Model, the Rate Belongs to the Fact Rather Than the Clock, and the Silent Change Only a Rate Could Catch Is the Case No Agent-Memory Benchmark Scores

(c) 2026 Pranay Mahendrakar. Licensed under CC BY 4.0. Language agents that keep memories across sessions usually discount a record by how long ago it was written or last read. That discount was introduced to decide what an agent attends to, but it is increasingly read as a statement about whether a record is still tru...

ToshLLM: local LLM inference on Intel Macs with AMD GPUs

A native SwiftUI application that runs large language models locally on Intel Macs with AMD GPUs, a configuration mainstream inference stacks leave unsupported or incorrect. Beyond packaging, it contributes original work to the Metal backend of llama.cpp: ToshGEMM, a manually tiled matrix multiply that replaces the sim...

SPIRIT-CONSORT-ELM: element-level annotated dataset and large language model approach for assessing randomized controlled trial reporting

Abstract Randomized controlled trials (RCTs) are central to assessing the benefits and harms of interventions, but incomplete reporting undermines their verifiability and usefulness. Although SPIRIT and CONSORT reporting guidelines promote complete reporting of RCT protocols and results publications, many RCTs remain i...

Causal Encyclopedia

Causal Encyclopedia — Cumulative PASS295 A Canonical Knowledge Infrastructure for LLM-Assisted Reasoning The Causal Encyclopedia is a cumulative, machine-oriented knowledge infrastructure designed primarily to support large language models, reasoning systems, and human–AI research workflows. Rather than functioning as...

AutoQA-MAS: An LLM-Based Multi-agent System for Autonomous Website End-to-End Testing

Manual execution of website end-to-end (E2E) test cases is a repetitive, labour-intensive process that creates bottlenecks in modern software development workflows. We investigate the feasibility of automating this process using AutoQA-MAS, a large language model (LLM)-based multi-agent system (MAS) comprising four spe...

IWAC Sentiment Analysis Visualization

Interactive SvelteKit dashboard for exploring large language model sentiment annotations of the Islam West Africa Collection (IWAC), a corpus of 12,349 francophone West African press articles from 57 newspapers, published between 1961 and 2025 in Benin, Burkina Faso, Côte d'Ivoire, Niger and Togo. Each article is annot...

Little Brains, Big Feats: Exploring Compact Language Models

NRITYAM: Language Models Meet Art and Heritage of Dance

Deploying LLM Inference on a Repurposed UMA APU: Transferable Lessons from a Vulkan-Only, 16 GB Edge Platform

Cost-driven interest in running large language models (LLMs) on non-mainstream silicon outpaces the maturity of the surrounding software stacks. This paper uses one such platform, the AMD BC-250 (a repurposed cryptocurrency-mining board with a GFX1013 "Cyan Skillfish" accelerated processing unit (APU), 16 GB of unified...

The Federico Maya Eternity Theorem: Maya Space, Maya Spectral Theory, and the conservation of information (v.37) with Analytical Supplement Lean 4.

The Federico Maya Eternity Theorem is one work in three offices. Maya Space is the place on which a configuration is read. Maya Spectral Theory is the spectrum of that place: a finite point spectrum, four atoms {9, 16, 21, 24} κ², and an essential spectrum beginning at 25κ². The conservation of information is the state...

What Gradients Add to Text Leakage in Split Language Models, Counted per Token and per Document

Split learning lets a client train a language model on a server without sending its text. The client runs the first layers itself and sends the server only their output, a vector of numbers for each token. During training, the server sends gradients back. We show that an observer at the split can rebuild most of the cl...