Loathsome Jargon

An AI Glossary for Genealogists and Family Historians

If you’d like to learn more about why I wrote this glossary of Loathsome Jargon, this January 7 2025 blog post explains.

The Loathsome Jargon Glossary:

Context Window (noun)

Context Window (noun): The short-term memory of an AI model, defining how much information it can process and “remember” during a single session. This includes all uploaded documents, your current prompt, and previous conversation exchanges. Exceeding this limit can lead to forgotten details or hallucinations (errors).

Example: In a 2023 genealogy class, Thomas Jefferson’s concise three-page will fit neatly within the AI’s context window, ensuring accurate results. By contrast, George Washington’s six-page will exceeded the capacity, risking omissions or errors. While today’s models (2025) have expanded windows, such as GPT-4o with 128,000 tokens and Gemini with 1,000,000, respecting their limits remains essential.

Practical Tips:

  1. Stick to 25% of the model’s capacity for optimal reliability. For GPT-4o, this is about 48 pages, and for Gemini, 375 pages.
  2. Be concise in your prompts—include only the most relevant details.
  3. Monitor the length of ongoing conversations, as older exchanges may fall out of the context window.

Why It Matters: A well-managed context window minimizes errors and enhances the AI’s ability to process genealogical data accurately, making it an indispensable concept for researchers to grasp.

Jagged Frontier (noun)

Our first bit of jargon comes from Prof. Ethan Mollick of Wharton, who wisely reminds us: “AI is weird.”1 Large language models (LLMs) are amazing, baffling, and often infuriating—a double-edged sword of potential and pitfalls. They make things up (“hallucinations”), give inconsistent answers (“indeterminate”), and remain mysterious even to their creators (“mechanistic interpretability” awaits us later). Above all, they’re unpredictable.

Mollick explains, “AI is weird. No one fully understands its capabilities, failures, or best uses. Some complex tasks (e.g., idea generation) are easy for AI, while simpler ones (e.g., basic math) are hard.” This uneven ability creates a “jagged frontier”—a shifting boundary between what AI excels at and where it fails.2

This frontier isn’t static—it’s amoebic, constantly expanding with each model update. What fails today might succeed tomorrow. For genealogists, this means:

  1. AI’s quirks are normal—don’t despair.
  2. Successes and failures vary even for similar tasks.
  3. Today’s limits often become tomorrow’s breakthroughs.

What can you do now? Track your failures. Not only will it save time by highlighting what doesn’t work, but it creates a ready list of test cases for future models. What fails with GPT-4 might thrive with GPT-5. AI’s frontier is jagged, but it’s growing—map it, and you’ll grow with it.

Jagged Frontier of AI Task Capabilities: An uneven boundary showing tasks AI performs well, struggles with, or fails to complete, adapted by Steve Little from an idea by Ethan Mollick.

This entry was first published in a longer form here: “Loathsome Jargon: An AI Glossary for Genealogists,” https://aigenealogyinsights.com/2025/01/07/loathsome-jargon-an-ai-glossary-for-genealogists/.

Sources:

  1. Ethan Mollick, “Centaurs and Cyborgs on the Jagged Frontier,” September 16, 2023, available at https://www.oneusefulthing.org/p/centaurs-and-cyborgs-on-the-jagged, accessed January 7, 2025. ↩︎
  2. Ethan Mollick, et al, “Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality” Harvard Business School Technology & Operations Management Unit Working Paper No 24-013, The Wharton School Research Paper, September 15, 2023, available at http://dx.doi.org/10.2139/ssrn.4573321, page 4, accessed January 7, 2025. ↩︎