Joanna Hofer

Project Calendar

AI robot points at a wall calendar for July 2025

Image by ChatGPT 4

This page shows a record of progress on the 30-day challenge to develop a grammar and vocabulary reference book for the Todos Santos Cuchumatán dialect of Mam using AI.

  • Day 1: Thursday, July 17, 2025 – Began working with ChatGPT 4.0; discovered, defined, and set up Truth and Respect Modes; set up designated GPT; gathered language resources; created AI quickstart guides to curb hallucinations and sycophancy.
  • Day 2: Friday, July 18, 2025 – Began building an organizational framework to collect language data; began working in Google drive, Google Docs; began backing up work; lost 50 percent of the work as a result of GPT Linguist crashing; decided not to use a GPT again. (LEARNING: GPTs can be unstable, don’t offer a folder system, and don’t offer any advantages for this kind of rigorous work.)
  • Day 3: Saturday, July 19, 2025 – Regrouped after data loss; interogated ChatGPT on how best to work on a project this complex; decided to explore more available options before committing to a particular AI platform; wrote “interview” script to probe the capabilities of other AIs; developed webpages where I could track my progress and make it visible.
  • Day 4: Sunday, July 20, 2025 – Reorganized files and took a file/source inventory; created and built out project donmain supports, including writing and finding images for Backstory, Interviewing AIs, Creating/Maintaining the Calendar, Gathering Sources and and Tools, Identifying Important Prompts/ Prompt Strategies/Prompt Styles ; found map of Todos Santos Cuchumatán, added attributions to Google Maps; (before publishing to a book format will need to create a map using a map creating program like mapbox.com, openstreetmap or umap); purchased Mam grammar book from Abe’s Books; recorded sources (with links) and resources on Sources page.
  • Day 5: Monday, July 21, 2025 – Worked with ChatGPT 4 to refine interview script and also tailor script to each AI candidate; interviewed: ChatGPT 4, Grok, Perplexity, Claude, Gemini, Character AI, and recorded their responses; (LEARNING: ChatGPT4 is an excellent resource for crafting prompts aimed at other AI platforms; many learnings here.); discussed the idea of using all these AIs as a team in roles that maximize their inherent strengths; conceived of final output document as a practical teaching tool with language acquisition results being the main goal; discussed using Character AIs and a GPT as student learners to test teaching materials; discussed best way to use command words to ensure Truth, Respect, and Intellectual Hygiene; (LEARNING: ChatGPT confirmed it would be more effective if I pasted in a complete bulleted list of communication expections upon initiating our chats.); Revisited my goal: decided I want to create a Todos Santos Conversation Guide that can be published on Amazon KDP; while I want to gather materials to document a more detailed and comprehensive guide for Todos Santos Mam, it isn’t my ultimate goal to produce an academic work; instead I want to produce a practical guide that can offer immediate help to communities; with help from ChatGPT identified Todos Santos Mam Scholars that I would request reviews from; updated all webpages as I progressed through the day’s work.
  • Day 6: Tuesday, July 22, 2025 – find a YouTube channel that explains how to use a phonemic chart and create a phonemic chart especially for Todos Santos Mam; track conversations with AIs and record any anomolies; ensure source materials used to generate the Todos Santos Mam phonemic chart are correctly representing Todos Santos Mam; add links to sources for easy referencing. Learning: ChatGPT4 admitted to trusting it’s own judgment over mine when making a decision for how to refer to Todos Santos Mam and so repeated an error after I gave instructions to remove the error.
  • Day 7: Wednesday, July 23, 2025 – Reviewed and analyzed ChatGPT recurrent hallucinations using Claude Sonet 4; Reviewed interview data, determined strengths/weaknesses of each AI for the Assistant Linguist role; asked an Claude Sonnet 4 to process the data and return a succinct summary; enlisted Claude to graph the results for easy visual comprehension; identified AIs specifically geared to linguistics; worked with ChatGPT 4 to structure book; Learnings: To Claude, a leading question is a question written without constraints; for example: what is the indigenous name for…? Instead ask, Is there a documented indigenous name for…? Got ChatGPT to generate a test for me to refine my prompt writing so I can avoid hallucinations; rescoped the book from a grammar book to a Mam conversation book.
  • Day 8: Thursday, July 24, 2025 – Write TOC and identify chapters (contents) to cover; generate illustrations for each chapter using ChatGPT/DALL-E.
  • Day 9: Friday, July 25, 2025 – Day Off.
  • Day 10: Saturday, July 26, 2025 – generate illustrations for each section (relying on DALL-E). Start developing a list of potential reviewers for the book.
  • Day 11: Sunday, July 27, 2025 – Identifed Mam words that are related to my pre-identifed topics and began building a Spanish/Mam/English vocabulary; developed related illustrations using ChatGPT and DALL-E. Used the YouTube videos made by Isa Pablo as my primary reference; will check everything with my grammar books later; possibly will incorporate IPA symbols later or a QR code that links to Isa Pablo’s videos or to vocaroo recordings. Key decision: This guide will be enormous if I don’t break it up by section. Trying to decide how to do that. Body parts and medical will be first; this could be helpful for school nurses and parents.
  • Day 12: Monday, July 28, 2025 – Finished reviewing second body part video by Isa Pablo and made vocabulary pages related to medical and body parts; developed illustrations with ChatGPT (DALL-E).
  • Day 13: Tuesday, July 29, 2025 – Worked on identifying the Todos Santos Mam alphabet using the video from Isa Pablo and the INALI resource. [This was difficult because there is a standard alphabet established by ALMG: Academia de Lenguas Mayas de Guatemala that covers all Mayan languages and dialects, but not all letters are used by Todos Santos Mam and there is some disagreement among sources as to which letters are used. I used INALI as my primary resource and Isa Pablo’s YouTube video on the alphabet as my secondary resource. My plan now is to go back and read through the grammar books, which don’t list the alphabet and deduce what the alphabet is. I’m not confident that I will come up with a definitive answer based on the information I have.] Learning: ChatGPT cannot accurate “read” a PDF and extract vocabulary because of how PDFs are rendered; however, I can take a screen shot of a page in a PDF and have ChatGPT extract words from the page. This is useful for developing vocabulary lists without haveing to type everything in myself. I only need to check that ChatGPT hasn’t made any errors.
  • Day 14: Wednesday, July 30, 2025 – Compiled information to develop the Todos Santos Mam alphabet listing and add IPA symbols
  • Day 15: Thursday, July 31, 2025 —day off
  • Day 16: Friday, August 1, 2025 — developed a chart to compare different versions of the alphabet; created dentist office and tooth images.
  • Day 17: Saturday, August 2, 2025 — began extracting vocabulary from textbooks; I want to see if AI can extract vocabulary words (correctly) from YouTube videos; Rewrite KEY Truth, Respect, Intellectual Hygiene, and Citation protocols as bulleted lists to ensure AIs are following my instructions exactly at the outset of every session; develop a list of potential reviewers for the book; follow up on tips given to me by Grok and Perplexity during their interviews; follow up on the lack of parallel review provided by Claude.
Back to Mam grammar