Andrew Marble

Misc writing

LLM System Data Exfiltration is not an LLM Issue, Aug 25, 2024

AI cybersecurity lessons from genAI, June 21, 2024

You won't train a better Gen AI model from your desk, April 13, 2024

Hyperbole in LLM testing, April 7, 2024

Data bias is giving way to training bias, Feb 24, 2024

Accountable large language models work by storing labeling effort. Prompting is not a substitute. Jan 26, 2024

AI Trends for 2024, Dec 30, 2023

Large Language Model Inference in Fortran, Dec 27, 2023

Private indemnification of AI models against copyright violation is a symptom of government inaction, Oct 13, 2023

AI Model Weight Providers Should Not Police Uses, No Matter How Awful They Are, Aug 30, 2023

Prompting, realized, and unrealized bias in generatvie AI, Aug 20, 2023

Why host your own llm? Aug 13, 2023

llama2.ipynb (not really writing, it's from scratch language model inference, with commentary, to try and show what the AI hype is really about). July 30, 2023 Also on github

Technical considerations for applying copyright and copyleft to AI, July 25, 2023

Licenses used for AI model weights, July 21, 2023

Foundational AI models do not violate copyright, July 18, 2023

Copyright for AI Model Weights, July 10, 2023

Motivations and strategy for open sourcing AI model weights, July 7, 2023

Software licenses masquerading as open source, June 1, 2023

Using Gnuplot for ascii charts inside notebooks, May 17, 2023

Into the great wide open AI, May 10, 2023

Content and controversy in generative AI models, April 21, 2023

Random numbers in bash, March 2023

Tempering expectations for generative AI, March 31, 2023

The need for open AI models, Feb 21, 2023

Moving from AI use-cases to products, January 24, 2023

Manufacturing is leading the way in accountable AI, November 22, 2022

Getting more out of experts by focusing on results and not process: observations of people and neural networks, May 17, 2021

Recent academic work

Computer vision defect detection on unseen backgrounds for manufacturing inspection. Expert Systems with Applications (2024)

Recent Software

llm.f90: fast, hackable large language model inference in Fortran, runs Meta's llama2 and variants as well as Mamba state space models

ferrite: transformer embedding model in pure Fortran, generates sentence embeddings using sentencetransformers models

Other

Resume

Github

Available for consulting, speaking engagements, coffee/beer, etc. See resume for contact info.