Large-Language-Models

Diagram showing text, molecular structures, and reactions feeding into a multimodal index and search system that outputs passages with context

Multimodal Search in Chemical Documents and Reactions

This paper presents a multimodal search system that facilitates passage-level retrieval of chemical reactions and molecular structures by linking diagrams, text, and reaction records extracted from scientific PDFs.

Computational Chemistry

ChemVLM architecture showing molecular structure and text inputs flowing through vision encoder and language model into multimodal LLM for chemical reasoning

ChemVLM: A Multimodal Large Language Model for Chemistry

A 2025 AAAI paper introducing ChemVLM, a domain-specific multimodal LLM (26B parameters). It achieves state-of-the-art performance on chemical OCR, reasoning benchmarks, and molecular understanding tasks by combining vision and language models trained on curated chemistry data.

Document Processing

Stream accuracy versus relative throughput for Mistral-7B and XGBoost models

LLMs for Insurance Document Automation

We explore LLM applications for page stream segmentation in insurance document processing, demonstrating that parameter-efficient fine-tuning achieves strong accuracy but revealing significant calibration challenges that limit deployment confidence.

Document Processing

Diagram showing page stream segmentation workflow: an input stream of pages is processed through binary classification of page pairs to predict document breaks, producing segmented output documents

LLMs for Page Stream Segmentation

We create TabMe++, an enhanced page stream segmentation benchmark with commercial-grade OCR, and show that parameter-efficiently fine-tuned decoder-based LLMs like Mistral-7B achieve 80% straight-through processing rates, dramatically outperforming encoder-based models.

Natural Language Processing

A nonsensical trigger sequence 'WTC theoriesclimate Flat Hubbard Principle' is fed into GPT-2, which then generates Flat Earth conspiracy text

GPT-2 Susceptibility to Universal Adversarial Triggers

We demonstrate that universal adversarial triggers can control both the topic and stance of GPT-2’s generated text, revealing security vulnerabilities in deployed language models and proposing constructive applications for bias auditing.