I am honored to be speaking on a panel this week on AI in the FOIA Process at the Virginia Open Government Coalition's annual conference. In preparation for this conference, I put together this list of resources for people unfamiliar with some newer concepts in AI and how they are being used today.

Technology Overview:

1. What are text embeddings?

2. What is semantic search?

3. An Overview of LLMs and General Applications

4. Retrieval Augmented Generation (RAG) - Used in most current chatbots / Q&A systems

AI in Practice

Some recent-ish stories outlines the possibilities and pitfalls of these technologies, for companies big and small.

1. Casetext used early access to GPT4 to build a product eventually acquired by Thompson-Reuters for $650 million

2. US Department of State reported on three separate pilots of AI tools in the FOIA process, one showing that automated review results in many false positives for redaction

3. Harvard’s Caselaw Access Project uses AI tools to increase public access to archival court records

4. Increasing case law access may be leading to a legal research boom

5. Oops, Air Canada found liable for its chatbot’s bad advice to a grieving grandson

6. Google Gemini generated racially-diverse Nazi soldiers, falling into the same issue as Dall-e did in 2022 while trying to account for bias in the model

Deep Dives on Current Problems (academic papers)

These are not for the faint of heart, but included for any who want a deeper understanding of current problems in this space and the current state of research in addressing them.

1. Survey on Hallucinations

2. Bias in LLMs

3. Alignment and Safety