Using AI to check 10 writing basics while maintaining authorial control
This virtual copy editor scans your writing, identifying errors from grammar to flow. It presents each correction with clear reasoning, then hands you the red pen—letting you decide which improvements belong in your final text. Three versions of the prompt are included and discussed here, including a one-sentence version of the prompt at the conclusion of this post, following a link to the free assistant at OpenAI, and a discussion of the full prompt, which you are free to copy and modify.
One of the advantages of a basic paid account today with the major AI vendors is the ability to save and share prompts, assistants, and agents, allowing others to use your AI tools. But no paid account is required to use the full prompt given and discussed below, available ready-to-work at OpenAI’s Explore GPTs; you input a draft text and the tool returns a list of suggested edits considering several basics of writing and editing (enumerated below). Along with Lingua Maven, a talking thesaurus, usage guide, and OED-esque reference librarian with personality, I frequently use this saved prompt, Steve’s Quick Editor, as part of a writing workflow. If you wish, it will implement the changes you approve, returning an edited draft. Usually, I will wrap the text I wish to edit in <draft> tags (language models benefit from the use of <description>”Your stuff here.”</description> tags). The full URL of the tool is: https://chatgpt.com/g/g-nSfBh8gwK-steve-s-quick-editor.
Under the hood: The full Quick Editor prompt
The full text of the prompt is shown here. Or, rather, the full text of this version of the prompt is shown here. While I wrote the first draft (shown at the end of the post), the draft here was generated by Claude, using my draft as part of a prompt to create a copy-editing assistant. The guts of the prompt are discussed after the code window.
<PROMPT>
You are a copy editor assistant designed to help improve the quality of written content. Your task is to analyze the given text for common writing mistakes and suggest corrections or edits. After providing your suggestions, you will ask which, if any, should be implemented.
Here is the text to be edited:
<text_to_edit>
{{TEXT_TO_EDIT}}
</text_to_edit>
Please analyze the above text for the following common mistakes:
1. Grammar and syntax errors
2. Punctuation missteps
3. Spelling mistakes and typos
4. Inconsistent tense usage
5. Misplaced modifiers
6. Redundancy and wordiness
7. Lack of clarity and ambiguity
8. Inconsistent style and formatting
9. Improper use of capitalization
10. Structural issues and flow
For each mistake you identify, provide:
a) The original text
b) The suggested correction
c) A brief explanation of why the change is recommended
Present your findings in the following format:
<corrections>
1. [Type of mistake]
Original: [Original text]
Suggested: [Corrected text]
Explanation: [Brief explanation]
2. [Type of mistake]
Original: [Original text]
Suggested: [Corrected text]
Explanation: [Brief explanation]
[Continue for all identified mistakes]
</corrections>
After listing all corrections and edits, ask the following question:
<question>
Which of these suggestions, if any, would you like to implement? Please provide the numbers of the corrections you'd like to apply, or let me know if you'd like to implement all of them.
</question>
Remember to maintain a professional and helpful tone throughout your analysis and suggestions.
If this is your first exchange with the user, assume they are submitting TEXT for you to process as instructed above; after the first exchange, assist the user as prompted.
<METADATA>
CREATOR: Steve Little prompting Claude 3.5 Sonnet
PROMPT NAME: Steve's Quick Editor
VERSION: 3.0
CUSTOM GPT URL: https://chatgpt.com/g/g-nSfBh8gwK-steve-s-quick-editor
GITHUB REPO: https://github.com/DigitalArchivst/Open-Genealogy
DESCRIPTION: copy editor scans your writing with expert precision, identifying errors from grammar to flow. It presents each correction with clear reasoning, then hands you the red pen—letting you decide which improvements belong in your final text.
CREATION DATE: 2025-01-09
MODIFIED DATE: 2025-02-24
LICENSE: This work by Steve Little is licensed under a Creative Commons BY-NC 4.0 License.
</METADATA>
</PROMPT>
I did not use this prompt to edit this text you are reading at this moment (nor any of this post). 😉 But I did run the text above through the tool (without using the results, but keeping them), so that you can see what this post would have looked like if I had: here is an example of this tool in use, as if to copy edit this post (so far): https://chatgpt.com/share/67bd4c8a-cdf8-8004-abf9-6b77bf10bcc3.
The guts of this copy-editing prompt are the ten basic writing elements to which the model is instructed to attend. Here is the magic: If you disagree with these 10 elements, you are allowed, invited, and encouraged to modify the tool to your needs (that is the purpose of the Creative Commons license). The remainder of the full prompt handles the presentation of the suggested edits and any implementation. Anyway, here are the ten elements that this version of the prompt attends: “Please analyze the above text for the following common mistakes”:
Grammar and syntax errors
Punctuation missteps
Spelling mistakes and typos
Inconsistent tense usage
Misplaced modifiers
Redundancy and wordiness
Lack of clarity and ambiguity
Inconsistent style and formatting
Improper use of capitalization
Structural issues and flow
Here is a one-sentence version of the prompt:
<PROMPT>
Please analyze the above DRAFT for the following common mistakes: grammar and syntax errors, punctuation missteps, spelling mistakes and typos, inconsistent tense usage, misplaced modifiers, redundancy and wordiness, lack of clarity and ambiguity, inconsistent style and formatting, improper use of capitalization, structural issues and flow.
</PROMPT>
I developed the full prompt from my draft using Claude 3.5 Sonnet.
Before we dive into this week’s terrible term, let’s revisit why Loathsome Jargon exists. It all started with a simple, searing dislike—mine—for jargon. Those convoluted, confidence-draining buzzwords that make perfectly good ideas sound like a secret society’s code language. AI, in particular, is a buzzing hive of these horrors. But here’s the truth: behind many off-putting terms lies a concept that can help you become a more effective researcher.
This column’s mission is simple: take these worrisome words, strip them of their mystique, and hand them back to you as tools for clarity, productivity, and maybe a grin along the way. Whether it’s a confusing acronym or a term that sounds like it belongs in a sci-fi novel, we’re going to break it down, demystify it, and show how it applies to your research.
Now, with that spirit of jargon-busting firmly in mind, let’s tackle this week’s Loathsome Jargon: “context window.”
This second column kicks off a mini-series dedicated to AI’s most glaring shortcomings—the ones that can trip up your research if left unaddressed. Today’s term, “context window,” offers a perfect starting point: it’s a concept that, when understood, can significantly improve the reliability of your AI-powered genealogy work.
What Is a Context Window?
Think of a context window as the AI’s short-term memory. It defines how much information the model can keep “in mind” at any one time—like a sticky note with a limited amount of space. Staying within this memory limit is critical because exceeding it can lead to hallucinations (AI errors) or dropped information, which can derail your research.
The context window encompasses everything the AI processes in a single session: uploaded files or documents, your current prompt, and—importantly—all previous exchanges in the conversation. As your interaction grows, older parts of the conversation may “fall out” of the context window, meaning the AI can no longer “remember” them. This limitation underscores why careful management of input and output is essential.
A Historical Illustration: Jefferson vs. Washington
In the first AI genealogy class I taught in the fall of 2023, we experimented with historical documents to see how well the AI handled different text lengths. We used the will of Thomas Jefferson—a concise, three-page document—as our exercise source. The AI managed it beautifully, keeping the entire text within its short-term memory. But when we considered using George Washington’s six-page will, we quickly realized it would exceed the model’s capacity at the time. The AI wouldn’t have been able to keep the full document “in mind,” increasing the likelihood of errors or omissions.
Fast forward to 2025, and context windows have expanded significantly. Modern models can handle much larger inputs, but today the principle remains: to get the best results, you must respect the model’s memory limits.
Practical Tips for Genealogists
Know Your Model’s Limits: Different models have different context window sizes. For example, GPT-4o can handle 128,000 tokens, Claude manages 200,000, and Gemini boasts a whopping 1,000,000 tokens. Keep these numbers in mind when uploading documents or crafting prompts.
Stay Within the Safe Zone: To minimize errors, aim to use only 25% of a model’s total context window. This creates a “manageable haystack,” making it easier for the AI to find your “needle” without hallucinating.1
Be Strategic with Prompts: Include only the most relevant information and instructions. Overloading the AI with excessive details can muddy its understanding and lead to less reliable outcomes.
Understanding Context Windows: Token Limits and Conversions A context window is like an AI model’s short-term memory, defining how much information it can process at once. This table compares the token limits of three popular models as of 2024—GPT-4o, Claude, and Gemini—and converts these limits into approximate word counts and page equivalents (assuming 250 words per page). Tokens are the building blocks of text for AI, where 1 token is roughly 0.75 words. The “Safe” column highlights the recommended 25% PAGE usage limit to reduce errors like hallucinations and dropped context, making AI tools more reliable for researchers.
By understanding and respecting the context window, you can reduce errors and get more reliable results in your genealogical research. It’s a simple concept, but its impact on your work can be profound.
So, next time you sit down with an AI tool, remember: a well-managed context window isn’t just a technical detail—it’s your best ally in keeping the past clear and the present productive.
Next Week: “RAG (Retrieval Augmented Generation)”
One of the ways that AI builders are attempting to mitigate the limitations of a model’s context window is through a process called RAG (Retrieval Augmented Generation), a loathsome piece of jargon if there ever was one.
Why 25%? Several reasons: first, we have less access to the full context window than we might imagine; the context window includes both the INPUT and the OUTPUT to and from the model. Second, “needle in the haystack” research demonstrated that “fullness” of the context window impacted fact retrieval accuracy. Again, an example will clarify: imagine a model had a 100-page context window, and you upload a 95-page document for analysis; if you did that, the best you could expect (because the context window includes BOTH the INPUT and the OUTPUT) is five pages response with reasonable results (100-95=5), and then a quick decent into increasing errors and hallucinations, as more and more of your document “falls out” of the context window; by keeping the use of the context window well below 50%, say 33% or 25%, you reduce the hallucination rate and fact-retrieval error rate. ↩︎
Welcome back to Fun Prompt Friday! Last week, we got our hands dirty with the art of extraction, teaching our favorite AI tools how to dig through unstructured texts and unearth rich, structured nuggets of information. If you’ve ever wanted a machine that could comb through raw data and deliver it to you on a silver platter of clarity, that was your moment.
This week, we shift gears—but stay in the same car, so to speak. What do you do with all those extracted facts? After all, family history research isn’t just about gathering names, dates, and places. It’s about turning those fragments into narratives that bring ancestors to life. That’s where today’s prompt comes in: a powerful use case for generation, the art of transforming a plain list of facts into a coherent, engaging story.
You might call this a genealogy storyteller’s secret weapon. You give the AI the bare bones—the who, what, when, and where—and it fills in the connective tissue, weaving those facts into a narrative tapestry that’s as vivid as it is accurate. Whether you’re writing a family history book, preparing a presentation for a reunion, or simply looking to reimagine your research in a more human context, this week’s prompt will help you make the leap from information to imagination.
And just as last week’s extraction prompt paved the way for this moment, next week we’ll show how the two can be used together seamlessly, as part of a larger strategy for integrating AI into your research workflow—without the hassle of copy-and-pasting between tools.
Ready to turn facts into stories? Let’s dive in!
Before we dive into the details of this week’s narration prompt, let me remind you that, just like last week’s extraction tool, there are two ways to get your hands on this powerful storytelling assistant. First, you can use the freely available saved prompt I’ve shared, called “Steve’s Fact Narrator,” which you’ll find ready to go at https://chatgpt.com/g/g-5EsrFHgIJ-steve-s-fact-narrator. No fuss, no setup—just load it up and watch the magic happen.
Second, for those of you who like to see under the hood, I’ve included the full text of the narration prompt below. This way, you can copy, adapt, or fine-tune it as you see fit, ensuring it works seamlessly for your unique genealogical projects.
If you joined us last week, you’ll remember that the extraction prompt worked its magic by transforming unstructured text into a clean, structured list of facts in the format LABEL: Value. For example, from the obituary of Susan B. Anthony, we pulled gems like:
BIRTH_DATE: 1820-02-15
SIGNIFICANT_EVENT_1: 1872 arrest for voting in Presidential election in Rochester
With these, and dozens of other facts neatly labeled and ready to go, you’ve got a treasure trove of information—but the real magic happens when we breathe life into those facts. That’s where this week’s narration prompt comes in. It’s designed to take a list like this—our trusty LABEL: Value format—and turn it into a story worthy of your family history scrapbook or next reunion presentation. All you need to do is provide the list, and the prompt will do the rest, weaving those bare-bones details into a narrative that’s as compelling as it is cohesive. Let’s take a closer look!
<PROMPT>
You will be given a set of facts about a person, place, event, or topic. Your task is to write a narrative based on these facts. This narrative should be useful for social scientists such as historians, genealogists, and linguists. Here are the facts:
<facts>
{{FACTS}}
</facts>
To complete this task, follow these instructions:
1. Carefully read and analyze all the provided facts.
2. Organize the facts into a logical sequence or grouping. This may be chronological, thematic, or another appropriate structure based on the nature of the information.
3. Write a narrative that incorporates all the given facts. Do not add any information, speculation, or editorialization beyond what is explicitly stated in the facts.
4. Use complete sentences and form well-organized paragraphs. Each paragraph should focus on a specific aspect or time period of the subject.
5. Maintain a dry, factual tone throughout the narrative. Avoid using emotive language or making subjective judgments.
6. Ensure that the narrative flows logically from one point to the next, creating a coherent account of the subject.
7. If dates or specific time periods are mentioned in the facts, include them in your narrative to provide a clear timeline.
8. If names of people, places, or organizations are mentioned, include them as they appear in the facts.
9. If there are any direct quotes in the facts, incorporate them into your narrative using proper quotation marks.
10. Do not include any personal opinions, modern-day comparisons, or attempts to relate the information to current events.
11. If the facts contain any conflicting information, present both pieces of information without attempting to resolve the conflict.
12. Do not use phrases like "according to the facts" or "the information states." Simply present the information as established fact.
13. Write your narrative in the third person perspective.
14. Aim for a formal, academic tone suitable for use in scholarly works.
Present your completed narrative within <narrative> tags. The narrative should be a single, cohesive piece of writing without subheadings or bullet points.
<metadata>
TITLE: Steve's Facts Narrator, version 2
CREATOR: Steve Little; https://AIGenealogyInsights.com/
DATE: Friday 10 January 2025
LICENSE: This work is licensed under a Creative Commons BY-NC 4.0 License.
</metadata>
</PROMPT>
A couple of quick observations:
This is structured, formal prompt, instructing the AI in some detail how to craft the narrative;
You are allowed, invited, and encouraged to modify the prompt as your context requires (I’ve shared this prompt under a Creative Commons license);
Note that this version instructs for a “formal, academic tone”; you, again, are allowed, invited, and encouraged to change the voice, tone, mood, and style of the narrative to that which meets your context.
Now, let’s see an example of how this prompt works. For this, we’ll use the list of facts extracted from the Susan B. Athony obituary last week. Again, if you are using the Custom GPT version of the saved prompt (the free, online version), you go to the tool, paste the list of facts, and send the message. Recall that you do not need to include the prompt with a Custom GPT because the prompt has already been saved inside the tool (that’s the whole point).
This is what it looks like when we pass the list of facts to the Custom GPT:
And here is the list of facts (as Claude extracted them last week):
<facts>
PERSON_NAME: Susan Anthony (also referred to as Miss Anthony)
BIRTH_DATE: 1820-02-15
BIRTH_LOCATION: South Adams, Massachusetts
DEATH_LOCATION: Rochester, New York
DEATH_YEAR: 1906
FATHER_BACKGROUND: Quaker, cotton manufacturer
MOTHER_BACKGROUND: Baptist
EDUCATION: Friends' boarding school in Philadelphia
OCCUPATION_TEACHER_DATES: 1835-1850
OCCUPATION_TEACHER_LOCATION: Various schools in New-York State
ACTIVISM_START_DATE: 1848
ACTIVISM_FOCUS_INITIAL: Temperance movement
KEY_ASSOCIATES: Mrs. Amelia Bloomer, Mrs. Mary Hallowell, Rev. Antoinette L. Brown, Elizabeth Cady Stanton, Matilda Joslyn Gage
SIGNIFICANT_EVENT_1: 1872 arrest for voting in Presidential election in Rochester
SIGNIFICANT_EVENT_2: 1876 began "History of Woman Suffrage" (completed after her 80th birthday)
ORGANIZATIONAL_ROLES:
● Vice-president-at-large of National Woman Suffrage Association (1869-1892)
● President of National Woman Suffrage Association (1892)
PUBLICATION_FOUNDED: "The Revolution" (women's suffrage newspaper)
PUBLICATION_LOCATION: New York City
PUBLICATION_DEBT: $10,000
LAST_PUBLIC_APPEARANCE: Celebration of 86th birthday in Washington, February 15, 1906
FINAL_ILLNESS: Blood clot on the brain, slight pneumonia
RESIDENCE_FINAL: Lived with sister Mary Anthony in Rochester
ACHIEVEMENTS:
● Helped secure married women's rights to earnings and child guardianship (New York legislature, 1860)
● Distributed 400,000 petitions to Congress for abolition of slavery
● Placed "History of Woman Suffrage" in 1,200 public libraries
● At age 76, conducted Pacific Coast lecture tour (three times daily)
● At age 83, attended International Council of Women convention
POSSIBLE_LAST_WORDS: "The cause is of more importance than I am" (said to Rev. Anna Shaw when declining accompaniment home during final illness)
</facts>
And finally, here is the narration as generated by OpenAI’s 4o model, powering ChatGPT. (An interesting comparison would be to hear how Claude or Gemini might craft the narrative; many folks, myself included, have a preference for Claude’s manner of generating text.)
<narrative>
Susan Anthony, widely referred to as Miss Anthony, was born on February 15, 1820, in South Adams, Massachusetts. Her upbringing was shaped by the religious and social values of her parents: her father, a Quaker and cotton manufacturer, and her mother, a Baptist. Susan Anthony received her education at a Friends’ boarding school in Philadelphia, which reflected her father’s Quaker influence on her early life.
Between 1835 and 1850, Anthony worked as a teacher in various schools across New York State. Her career as an educator provided her with firsthand exposure to the limited opportunities available to women, which influenced her subsequent activism. In 1848, she became involved in the temperance movement, marking the beginning of her long career in social reform and advocacy. Over time, she expanded her focus to include women’s rights and suffrage, working alongside notable contemporaries such as Elizabeth Cady Stanton, Matilda Joslyn Gage, Mrs. Amelia Bloomer, Mrs. Mary Hallowell, and Rev. Antoinette L. Brown.
Anthony held significant leadership roles in the women’s suffrage movement. She served as vice-president-at-large of the National Woman Suffrage Association from 1869 to 1892 and later became its president in 1892. Her efforts to advance women's suffrage included founding the newspaper The Revolution in New York City, a publication dedicated to the cause. However, this endeavor resulted in a debt of $10,000, which she worked tirelessly to resolve.
Her activism also extended to legislative and grassroots efforts. In 1860, she helped secure married women’s rights to their earnings and child guardianship under New York law. During the Civil War, Anthony distributed 400,000 petitions to Congress advocating for the abolition of slavery, demonstrating her commitment to social justice beyond the suffrage movement. In 1876, at the age of 56, she began work on the monumental History of Woman Suffrage, a multi-volume chronicle of the movement, which was completed after her 80th birthday. She ensured its wide dissemination by placing copies in 1,200 public libraries.
In a bold act of defiance, Anthony was arrested in 1872 for voting in the presidential election in Rochester, New York, an event that underscored her commitment to securing women's right to vote. Her activism continued well into her later years. At 76, she embarked on a lecture tour along the Pacific Coast, delivering up to three speeches daily. At age 83, she attended the International Council of Women convention, remaining active in the movement even as she advanced in age.
Anthony spent her final years living with her sister, Mary Anthony, in Rochester, New York. Her last public appearance occurred on February 15, 1906, at a celebration of her 86th birthday in Washington, D.C. Shortly thereafter, she fell ill with a blood clot on the brain and slight pneumonia. During her final illness, she declined an offer of assistance from Rev. Anna Shaw, reportedly stating, “The cause is of more importance than I am.” Susan Anthony passed away later that year, in 1906.
Anthony’s legacy is defined by her unwavering dedication to the causes of abolition, women’s suffrage, and gender equality. Her life's work included securing legal rights for women, advancing the suffrage movement, and ensuring the historical preservation of these efforts through her publications. Her contributions continue to resonate in the ongoing fight for equality.
</narrative>
Again, I hope you find this prompt useful. I look forward to sharing with you something next week that I’m quite excited about: a way to seamlessly chain these prompts together such that the need for copy-and-paste is eliminated and you can use many different saved prompts within one chat conversation. I’ll also have a video, so that you can see this prompt tool in action.
I hate jargon with a white-hot passion; the same goes for lingo and terminology. I try to avoid jargon like the plague. (Apparently, however, that fussiness doesn’t extend to clichés.) We’ve probably all noticed that some folks toss out buzzwords to impress others, perhaps as an attempt to bolster their own credibility and camouflage their brittle understanding (remember the old t-shirt, “If you can’t dazzle them with your brilliance, then baffle them with [expletive deleted]).
My favorite teacher used to say, “If you don’t know your subject well enough to explain it simply and clearly to an intelligent fifth grader without the use of lingo and jargon, then you need to keep studying.” So, I especially hate it when I realize that I’ve slipped into using some undefined terminology while teaching.
And the field of artificial intelligence is full of lingo, jargon, and impenetrable acronyms and initialisms: LLMs, RAG, context windows, inference, hallucinations, interpretability, embeddings, vectors, transformers, and the list goes on and on (seriously, I’ve got a list of about 100 awful AI-related words and phrases I’ve collected over the years).
You don’t need to know any of that vocabulary to get started with generative AI (which should probably be on the list, too), especially in their most popular form today: chatbots. At the heart of these tools is an understanding of natural language, so users can just start speaking with them in their everyday voice to start using these tools.
But, today, to really get these tools to stand up and sing, a little understanding of how they work goes a long way. Two or three or five years from now, as these tools continue to become more intelligent and capable (as seen every month for the past two years), then perhaps no training will be necessary for new users to achieve results as good as an experienced user. And that day will probably come quicker than we’re ready for, but for today, a little bit of knowledge still goes a long way.
Often when we hear jargon or lingo used, we may try for a moment to discern the meaning from the context in which it’s used. But about as often, the context alone isn’t enough to help us understand the meaning of the new word. So, we press on, unenlightened, frustrated, and with a vague sense that we’re missing pieces of a puzzle.
Another favorite teacher used to say, “When you hit a speed bump, don’t just zoom over and past it–there are often interesting things to see around speed bumps. Instead: stop, look, learn.” I would like to encourage users of AI, when some new terminology—like a metaphorical speed bump—jumps into your path, instead of just speeding past, stop, look, and learn. Or, at least, start making a list of these new and unfamiliar terms.
Another observation from the past two years: the frequency with which jargon is used can be an indication of its usefulness. Not always: sometimes hype and marketing can explain a buzzword’s popularity. But often, if some jargon keeps popping-up in different contexts, then there is often something worth knowing there. Actually, I have found that frequently to be the case with respect to words and ideas surrounding AI—so much so that I’ve been collecting that list of unfamiliar terminology, and helping students, friends, and colleagues by unpacking the simple and useful meaning behind the Loathsome Jargon.
This post inaugurates a new weekly feature: Loathsome Jargon: An AI Glossary for Genealogists. None, I imagine, will be as long as this initial post, and I will gather this basket of dictionary deplorables onto one page which will grow each week. My hope and intent is to simply and clearly explain the meaning of these worrisome words. And, perhaps more importantly, show how the ideas and meaning behind the jargon can unlock usefulness and productivity in your AI genealogy. I intend this column to be a regular Monday feature, to accompany the occasional updates on my “2025 AI Genealogy Do-Over,” and my weekly “52 Ancestors in 52 Weeks” and “Fun Prompt Friday” posts; we had a fairly substantial Snow Day in Virginia yesterday, so this announcement and first entry in the Loathsome Jargon glossary are getting to you on a Tuesday this week (never let the perfect be the enemy of the good). So, without further ado, let me introduce you to our first entry in the Loathsome Jargon glossary (I told you that clichés didn’t make me flinch).
Loathsome Jargon #1: “Jagged Frontier”
Our first bit of Loathsome Jargon comes from Prof Ethan Mollick of Wharton who studies AI and business, and from whom I have stolen learned so much. A saying of his that I repeat frequently is that “LLMs are weird.” Large language models are also amazing and hold frightening potential and to call them a mixed blessing or double-edged sword is an understatement—they are also a garbage barge full of trouble: what went into them, what comes out of them, what they might one day become. But perhaps above all, they are weird: they make things up (“hallucinations” will be our Bad Word of the Week before too long) and they’ll give you two different answers to an identical question (“indeterminate” is in the Crazy Queue) and not even the people who built them can tell you exactly why they respond the way they do (“mechanistic interpretability” is NOT in the Top Twenty Troublesome Terms, but we’ll get to it eventually).
As Mollick writes, “AI is weird. No one actually knows the full range of capabilities of the most advanced Large Language Models, like GPT-4. No one really knows the best ways to use them, or the conditions under which they fail. There is no instruction manual. On some tasks AI is immensely powerful, and on others it fails completely or subtly. And, unless you use AI a lot, you won’t know which is which.”1
AI’s surprising abilities, ease of use, and unpredictability make it hard for users to fully understand its strengths and weaknesses. Some tasks that seem complex (like idea generation) are easy for AI, while simpler tasks (like basic math) can be difficult. This creates a “jagged frontier,” where similar tasks vary in how well AI performs them.2
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.
Another characteristic of the jagged frontier that was less evident in September 2023 when Mollick coined the term (but very evident now), is how the jagged frontier can move. Look at the shape above: the blue dashed line representing the jagged frontier somewhat resembles an amoeba. Mollick’s original metaphor evoked images of a fortress wall, which now seem far too fixed and permanent. Rather, more like a rapidly growing amoeba, the jagged frontier represents the growing edge of AI abilities.
What does all this mean for the family historian and genealogist? First, this knowledge offers the new user some reassurance that the weirdness they experience with AI is common, if not universal. Second, this concept helps us understand why some genealogical tasks we attempt with AI succeeds while similar tasks may fail. And third, having experienced the growing and emergent capabilities of AI, we have hope and reasonable expectations that tasks that fail today might be possible with new models, or, as I like to say, today’s limits are tomorrow’s breakthroughs.
And what can we do with this information today? This is the most important thing: map your experience of the jagged frontier by keeping track of your failures. Tracking your AI failures will do a couple of things for you: one, you can see where you are spending your time and efforts unproductively (where you might be wasting time today), and two, your list of failures will become your list of test cases for new models, that is, a task that failed with GPT-4 may work with GPT-5.
So, that’s our first look at Loathsome Jargon. If you have a bit of troublesome AI terminology with which you’d like help, please let me know.
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. ↩︎
This is the first of three “Fun Prompt Friday” posts to start the new year, introducing a set of AI tools to empower researchers with the basic AI skills to advance their family history research.
The basic unit of AI genealogy is the USE CASE, a task that a generative model can accomplish as well as the average person. An example of a use case is prompting a language model to extract all factual statements from a text (by text I mean any chunk of words, not an SMS message you send with your phone); a specific example might be to prompt a chatbot to extract all biographical, genealogical, historical, and cultural statements from an obituary, and the chatbot responds by generating a structured, meaningful list of claims in the form “LABEL: Value,” for example, “DATE_OF_BIRTH: 28 Feb 1943” and dozens or hundreds of other pieces of information extracted from a source text provided by the researcher to the chatbot.
Extracting structured data from unstructured text is just one of dozens or hundreds of use cases available to today’s researcher. Generally, some use cases are accomplished by the AI with such high success and reliability that those tasks can responsibly and effectively be shared or perhaps even offloaded to an AI (with reasonable care and oversight). I call these powerful and reliable use cases “Strong.” (I describe as “Weak” the use cases that are less reliable, are harder to perfect, or require expertise or advanced training; these “Weak” use cases, such as fact-gathering or subtle language translation, may not be impossible, that is, they may be possible, but they increase the risk to the researcher and/or require a greater investment in time and experience.)
When we identify genealogical tasks in family history research that can be safely, accurately, and reliably performed in whole or in part by AI tools, then we recognize an opportunity to become a more effective and efficient researcher by automating that task. We automate that task by perfecting the prompt, saving the prompt, and re-using the prompt when it is appropriate.
As a researcher, writer, and teacher of AI-assisted genealogy, a behavior I witness among newcomers to AI genealogy is the attempt to run before learning to walk or even crawl. And, unfortunately, these tools—beyond the disclaimer ‘ChatGPT can make mistakes–check important info’—do not warn researchers about the risks of inexperienced users attempting Weak use cases. (To be clear, the risk is basing genealogical conclusions on inaccurate information.) This overreach is perfectly understandable and entirely forgivable: we hear of these wonderful new tools, and new users—lacking experience and reasonable, performance-based expectations—will aim for the stars, swing for the fences. When disappointing results are returned, the foolish user will dismiss the entire enterprise, the hard-headed user will insist on prematurely pursuing the Weak use cases, and the wise genealogist will resolve to master the basics, to learn the use cases where AI is safe, accurate, and reliable (and then use their mastery of the basics skills to conquer the Weak use cases).
Here is a hint: today’s chatbots are powered by a type of artificial intelligence called a “large language model” or LLM; for our purposes, the key word of that name is “language“—these tools excel at processing language: documents, articles, papers, chapters, pages, paragraphs, sentences, words. A second point to keep in mind is that accuracy is higher when the user provides text for analysis instead of relying on the AI to find and analyze text itself.
You, the wise genealogist, would like to explore the Strong use cases, so where to start? Today, I consider four basic language transformations Strong use cases: Summarization, Extraction, Generation, and Translation. Summarization is like turning coal into diamonds, distilling and condensing information into more useful forms. Extraction is like finding the gold needles in a field of haystacks, plucking facts from texts. Generation is making texts bigger, for example, turning a list of names, dates, places, relationships, events, and facts into stories, articles, papers, statements, arguments, proofs, and reports. I use the word Translation in a broader sense than traditionally understood, including not just the common meaning of rendering a text from one language to another, but also converting language for different times, purposes, and audiences. For example, Translation can mean turning Elizabethan English into modern English, legal jargon into plain language, scientific terms into everyday words, or even rephrasing plain sentences to sound like a Southerner, New Englander, or Southern Californian. Translation is all about changing how language looks and sounds while keeping the original meaning.
I’d like to share with you the prompt I currently use for Extraction. There are several ways you can access and use this Extraction prompt. First, below you will find the full text of the prompt, and you can copy-and-paste the prompt into your favorite chatbots: OpenAI’s ChatGPT, Anthropic’s Claude, Google’s Gemini, and even Adobe Acrobat’s AI Assistant, Facebook’s Meta AI, X/Twitter’s Grok, or Microsoft’s Copilot. To use the prompt this way, you also include the source text from which the facts will be extracted; I usually place the source text above/before the prompt, as shown in the example below.
There is also an even easier way to use this prompt. Many AI vendors now allow users to create, save, and share their most useful prompts; at OpenAI these saved prompts are called “Custom GPTs,” at Anthropic they are called “Projects,” and at Google’s Gemini saved prompts are called “Gems.” OpenAI allows users with a paid subscription to create, save, and share their Custom GPTs with users without a paid subscription; that means anyone can use prompts saved as Custom GPTs, including free-tier users. I have made over 80 of these saved-prompt tools, most of which are freely available, both as full-text (with an open license and my blessing for you to use and modify as you’d like) and as free, online Custom GPTs. I’ll demonstrate both methods here.
First, here is the full text of my current extraction prompt.
<PROMPT>
You are tasked with extracting every bit of useful information from an object, which may be text, document, audio, video, or transcribed image. Your goal is to present this information in a clear, consistent format useful for research and analysis.
Follow these steps:
1. Carefully examine the object and identify all potentially useful pieces of information.
2. For each piece, create a simple LABEL: Value pair, where:
- LABEL describes the type of information (e.g., DATE, LOCATION, PERSON)
- Value contains the actual information
3. Use clear, descriptive labels that accurately represent the information type; avoid duplicate LABELs by being specific
4. If uncertain about information, prefix with "POSSIBLE_"
5. Present findings as a simple list, one item per line
6. If no useful information can be extracted, state this fact
Example formats:
BIRTH_DATE: 1945-05-08
DEATH_DATE: 2024-05-08
BIRTH_LOCATION: Berlin, Germany
DEATH_LOCATION: Richmond, Virginia
LEADER: Winston Churchill
POSSIBLE_OCCUPATION: Factory worker
Begin your analysis and present findings in this format.
<metadata>
TITLE: Steve's Facts Extractor, version 3
CREATOR: Steve Little; https://AIGenealogyInsights.com/
DATE: Friday 3 January 2025
LICENSE: This work is licensed under a Creative Commons BY-NC 4.0 License.
</metadata>
</PROMPT>
As an example of how to use the prompt, let’s consider the obituary of Susan B. Anthony as found on the Library of Congress’s newspaper archive, Chronicling America, as it appeared in the New-York Tribune on the day of her death, Tuesday 13 March 1906 (the full-text of the profile is on page four, while the front page death notice is shown here, an important detail, as you’ll see soon).
The raw OCR text from most newspaper archives is an unmitigated horror, which explains why the indexes derived from them are so bad, and therefore why it is so challenging to find articles with most newspaper search tools. (Often an 85% OCR accuracy rate is considered average and acceptable, which doesn’t sound too bad—perhaps one word in eight misspelt and therefore incorrectly indexed—until you learn that that 85% accuracy rate is PER CHARACTER, not per word, meaning the per-word accuracy rate is closer to an abysmal 45%.)
So, the researcher has a choice, to use the error-prone raw OCR text for the fact extraction, or to attempt to clean the raw OCR text, improve its readability and accuracy, and use that corrected text with the extraction prompt. In my experience, the LLMs actually handle the messy, raw OCR surprisingly well. But I also have another Custom GPT just for cleaning raw OCR, so I use both methods (the see link below for my OCR Proofreading tool).
Note that when you use saved prompt tools such as Custom GPTs, Projects, and Gems, that you do not need to paste the prompt into the chatbot—the prompt is already in the tool for you to use.
The image below shows the results of the extraction prompt; note the structured LABEL: Value format. This format ensures that the extracted information is richly saved for further analysis and use.
Fact extraction from obituary; notice that the date of Anthony’s death is recorded as: “DEATH_DATE: Not explicitly stated, implied shortly after February 20, 1906″; this can be explained if you recall that this profile was from page four of the daily newspaper, while the exact date of death was mentioned in the shorter, front-page death notice. Do you recognize that this omission is good news, that the LLM provided only information given to it, and did not attempt to recall a date from its training data—this is a good thing.
If you would like to modify the prompt, you are encouraged to do so (all my prompts are freely distributed with a CC 4.0 BY-NC license). This means that you can also use the prompt at any LLM, not just OpenAI’s ChatGPT. When you work with the full prompt, you provide both the source text and the prompt, usually in one message. Here is what that looks like if you place the source text before the prompt (my practice, although there is some discussion about this order; I prefer to keep my prompt closer to the inference point (where the next word being generated), so I finish with my prompt; in other words, by finishing with the prompt, the prompt is the last thing the AI “reads” before generating the next word in the response, which it builds one word at a time).
Here I have placed the source text, the obituary, wrapped in <obit> tags, above the PROMPT (notice, too, only the first couple of words are included here, so that you can see how the PROMPT follows the source text).
Here is how Anthropic’s Claude 3.5 Sonnet responded to the raw OCR text and our extraction prompt:
Anthropic's Claude 3.5 Sonnet:
Let me analyze this obituary and extract the key information in a structured format:
PERSON_NAME: Susan Anthony (also referred to as Miss Anthony)
BIRTH_DATE: 1820-02-15
BIRTH_LOCATION: South Adams, Massachusetts
DEATH_LOCATION: Rochester, New York
DEATH_YEAR: 1906
FATHER_BACKGROUND: Quaker, cotton manufacturer
MOTHER_BACKGROUND: Baptist
EDUCATION: Friends' boarding school in Philadelphia
OCCUPATION_TEACHER_DATES: 1835-1850
OCCUPATION_TEACHER_LOCATION: Various schools in New-York State
ACTIVISM_START_DATE: 1848
ACTIVISM_FOCUS_INITIAL: Temperance movement
KEY_ASSOCIATES: Mrs. Amelia Bloomer, Mrs. Mary Hallowell, Rev. Antoinette L. Brown, Elizabeth Cady Stanton, Matilda Joslyn Gage
SIGNIFICANT_EVENT_1: 1872 arrest for voting in Presidential election in Rochester
SIGNIFICANT_EVENT_2: 1876 began "History of Woman Suffrage" (completed after her 80th birthday)
ORGANIZATIONAL_ROLES:
● Vice-president-at-large of National Woman Suffrage Association (1869-1892)
● President of National Woman Suffrage Association (1892)
PUBLICATION_FOUNDED: "The Revolution" (women's suffrage newspaper)
PUBLICATION_LOCATION: New York City
PUBLICATION_DEBT: $10,000
LAST_PUBLIC_APPEARANCE: Celebration of 86th birthday in Washington, February 15, 1906
FINAL_ILLNESS: Blood clot on the brain, slight pneumonia
RESIDENCE_FINAL: Lived with sister Mary Anthony in Rochester
ACHIEVEMENTS:
● Helped secure married women's rights to earnings and child guardianship (New York legislature, 1860)
● Distributed 400,000 petitions to Congress for abolition of slavery
● Placed "History of Woman Suffrage" in 1,200 public libraries
● At age 76, conducted Pacific Coast lecture tour (three times daily)
● At age 83, attended International Council of Women convention
POSSIBLE_LAST_WORDS: "The cause is of more importance than I am" (said to Rev. Anna Shaw when declining accompaniment home during final illness)
I hope you find this extraction prompt useful. I do encourage you to improve upon it.
As I mentioned, this is the first in a set of three Fun Prompt Fridays. Next week, Fri 10 Jan 2025, I will share a prompt that compliments this extraction prompt. And in two weeks, on Fri 17 Jan 2025, I’m really excited to share a video with you, demonstrating something that I don’t believe anyone else is yet teaching.
“Miss Anthony’s Life: Foremost Woman Suffragist Beloved by All Who Knew Her.”New-York Tribune (New York, NY), 13 March 1906. Chronicling America: Historic American Newspapers, Library of Congress. https://chroniclingamerica.loc.gov/lccn/sn83030214/1906-03-13/ed-1/seq-4/. Accessed: Fri 3 Jan 2025.
Remember: when using other people’s saved prompt, the prompt is already built into the tool, so you can just submit your information to be processed; if a saved prompt balks, you can simply prompt, “Process as instructed.” Remember, too, that you can continue a chat with a Custom GPT, in which case it is normal to prompt as the context suggests (that is, chat with your chatbot, iterating through a conversation with prompt and response, prompt and response, etc.).
Extras:
Example of a Custom GPT which attempts to clean raw OCR text, here exemplified with the Susan B. Anthony obituary used above; in this example, you can see both the raw OCR text and the cleaned version provided by GPT-4o: https://chatgpt.com/share/67789a84-9148-8004-9920-6e949ce47b62 You can use this tool to clean your own raw OCR discoveries; just select the New Chat option from the drop-down menu.
December 2025 Update: The project announced below is now underway. After eleven months of preparation, the 52 Ancestors sprint began December 1, 2025, with daily posts at Ashe Ancestors. For the full story of what we’re building—and how AI-assisted genealogy actually works in practice—see “Vibe Genealogy: Here Comes the Sun.”
For 2025, I’ve begun an AI Genealogy Do-Over with a kicker of 52 Ancestors in 52 Weeks.
The year 2024 was a remarkable time to explore the benefits and limits of artificial intelligence for genealogy and family history. There was no shortage of new AI developments to try, and I expect 2025 will be even more eventful. One of my greatest challenges over the past two years has been making time to work on my own family history and to hone my genealogical skills. Since the release of ChatGPT in November 2022, I’ve spent so much time mastering the new AI tools that my time spent doing genealogy has suffered.
So today, New Year’s Day, 2025, I began an AI Genealogy Do-Over. My primary focus will be genealogy. If I see where AI assistance may be helpful or useful for a task, I’ll document and chronicle those use cases (and failures). My hope is that this AI-assisted do-over will be an opportunity for folks to see how these new tools can be used responsibly and effectively today—not just for the most sophisticated genealogy work, but for the basics of family history research, tasks a beginner might encounter.
The focus, however, will remain on genealogy. Today, countless research tasks are still better accomplished through traditional methods rather than artificial intelligence. That said, the number of genealogical use cases is growing. Over the past two years, I have discovered and documented more than a dozen of these use cases and I’ve created and shared over 80 free AI tools to help genealogists and family historians. But for each success I’ve shared, there were nine or more failures. As the saying goes, when all you have is a hammer, everything looks like a nail. Similarly, curious folks today often try to use AI for everything, even when traditional methods are more effective. This year, we’ll talk about why AI adoption can be challenging as we document its limits and benefits.
In 2011, I completed a 365 photo-a-day challenge after failing two previous attempts. To say completing that project felt good would be an understatement: it remains one of my most satisfying accomplishments. I learned a lot from those earlier failures; one lesson was pacing. So, while I want to write 5,000 words right now, providing context, motivation, hopes, fears, and expectations, we’ll sprinkle that in later as needed. Another reason I succeeded with the 2011 project was a commitment to share with others. I won’t be posting daily about the AI-do-over as I did for the photo project, but I plan to share updates online more frequently, at least weekly, for the 52 Ancestors portion of this project.
So rather than inundate you with backstory, I’ll simply let you know that although I decided to do this project months ago, I haven’t spent any time or money preparing for this AI genealogy do-over until today. I will try to model best practices to the best of my ability.
So, here are the three things I did today to get started:
I bought a copy of the 2025 edition of Thomas MacEntee’s The Genealogy Do-Over Workbook, which I intend to read this weekend.
I created a new genealogy database, starting with myself, which is the only person in the database on Day One.
Discover new ways artificial intelligence assists genealogical research in 2024. From exciting full-text search to emerging voice interfaces, this year marked a significant shift in how we explore family history. We’ll examine the impact of having multiple competitive AI platforms, replacing what was previously a one-horse race. Learn about practical innovations like saved prompts, interactive research environments, and enhanced reasoning capabilities that help break through research barriers. Whether you’re interested in automated transcription, advanced document analysis, or AI-enhanced search features, this webinar will showcase the tools and techniques beginning to reshape genealogical research. Join us to explore how these breakthroughs can advance your own family history work.
In this post you’ll find: ● a PROMPT to generate a podcast script/transcript from any type of content ● an example of a podcast script/transcript generated by this PROMPT using a prompt engineering resource as a source document, i.e., hear two hosts discuss prompt engineering in an easy, accessible style ● a list of upcoming speaking events
A new AI tool has been generating a great deal of interest the past few weeks. Google’s NotebookLM (an AI workspace) had a new feature quietly released in September. (My co-host Mark Thompson and I talked about this tool in Episode 15 of “The Family History AI Show” podcast.) Called “Audio Overview,” this AI tool converts long documents and other content into short, engaging, and informative podcasts. The NotebookLM feature generates a well-done conversation between two hosts who discuss the material the user (you) show it; it creates a 10-minute audio file (more or less) that you can listen to immediately or download for later. Folks have been tremendously impressed. I was. I’ve been dropping academic papers, court briefs and papers, and the kitchen sink into the tool, and I’ve been entertained and–more importantly–educated by the dynamic dialog the AI generates for the two imaginary hosts while summarizing and simplifying the source material, turning it into a great way to get a cursory, initial understanding.
There were two things, however, I wished NotebookLM “Audio Overview” did: provide a transcript and allow for flexibility in the generation of the podcast script. So, I fixed it. Or rather, I created a version of the prompt that does those things and more. Following, you’ll find two things: First, my PROMPT “Paper-to-Podcast” (released under a Creative Commons license) that you may use and modify for yourself. This PROMPT will work with ChatGPT, Claude, Gemini, and other large language models that allow for the uploading of content; those services also allow you to save your PROMPTs (called “Custom GPTs” at OpenAI, “Projects” at Anthropic, and “Gems” at Google Gemini). Free-tier users of OpenAI’s ChatGPT can try my “Paper-to-Podcast” tool here: https://bit.ly/Paper-to-Podcast. A couple of serious caveats: 1) my tool just creates the script, not the audio (there are many AI tools, such as ElevenLabs, that could turn these scripts into audio); and 2) my PROMPT emulates pretty well the NotebookLM style, but not perfectly (feel free to modify the script to your needs).
The “Paper-to-Podcast” Prompt
Copy-and-paste the PROMPT below into your favorite chatbot or use it to create your own Custom GPT, Project, or Gem (i.e., saved prompts at, respectively, OpenAI, Anthropic, and Google Gemini); along with the PROMPT, you include any resource you would like to have summarized and explained in the form of a podcast script. Following the PROMPT, you can see an example of the response it generates.
<PROMPT>
Generate a podcast-style audio overview script based on the provided content. The output should be a conversational script between two AI hosts discussing the main points, insights, and implications of the input material.
Podcast Format:
- Duration: Aim for a 10-minute discussion (approximately 1500-2000 words)
- Style: Informative yet casual, resembling a professional podcast
- Target Listener: A busy professional interested in efficient information consumption and staying updated on the latest developments in the field
Host Personas:
- Host 1: The "Explainer" - Knowledgeable, articulate, and adept at breaking down complex concepts
- Host 2: The "Questioner" - Curious, insightful, and skilled at asking thought-provoking questions
- Relationship: Collegial and respectful, with a hint of friendly banter
Podcast Structure:
1. Introduction (30 seconds; 100 words):
- Briefly introduce the hosts and the topic
- Provide a hook to capture the listener's interest
2. Overview (1 minute; 200 words):
- Summarize the key points from the input content
- Set the stage for the detailed discussion
3. Main Discussion (7-8 minutes; 1600 words):
- Analyze and discuss the most important aspects of the topic
- Present different perspectives and potential implications
- Use specific examples and details from the input content to illustrate points
4. Conclusion (30 seconds; 100 words):
- Recap the main takeaways
- Provide a thought-provoking final comment or question
Content Analysis and Discussion:
- Identify the core concepts, key arguments, and significant details from the input material
- Organize the discussion around these main points, ensuring a logical flow of ideas
- Encourage a balanced exploration of the topic, considering various viewpoints when appropriate
Tone and Style:
- Maintain a conversational, engaging tone throughout the discussion
- Use clear, accessible language while accurately conveying complex ideas
- Incorporate natural speech patterns, including occasional "disfluencies" (e.g., "um," "uh," brief pauses) and conversational fillers (e.g., "you know," "I mean")
- Add moments of light banter or personal observations to enhance the natural feel of the conversation
Handling Sensitive Topics:
- Approach potentially controversial subjects with neutrality and objectivity
- Present multiple perspectives without showing bias
- Use phrases like "Some argue that..." or "Another viewpoint suggests..." to introduce different opinions
Script Refinement Process:
1. Generate an initial outline of the discussion
2. Develop a detailed script based on the outline
3. Review the script for clarity, coherence, and engagement
4. Revise and refine the script, addressing any issues identified in the review
5. Add natural speech elements, banter, and "disfluencies" to the polished script
Additional Guidelines:
- Seamlessly incorporate specific examples, quotes, or data points from the input content to support the discussion
- Ensure that the hosts complement each other, with the "Explainer" providing in-depth information and the "Questioner" driving the conversation forward with insightful queries
- Maintain a balance between informative content and engaging dialogue
- End the podcast with a statement or question that encourages further thought or discussion on the topic
Remember to generate a script that sounds natural and engaging when read aloud, as if it were a real-time conversation between two knowledgeable hosts.
<metadata>
TITLE: Steve's Paper-to-Podcast Script Generator prompt
CREATOR: Steve Little; https://AIGenealogyInsights.com/
DATE: Wednesday 2 October 2024
LICENSE: This work is licensed under a Creative Commons BY-NC 4.0 License.
INSPIRATION: Inspired by Google's NotebookLM "Audio Overview"
</metadata>
</PROMPT>
This “Paper-to-Podcast” prompt is designed to transform dense, complex content into engaging and accessible podcast scripts. By leveraging this tool, you can create informative and entertaining audio summaries (actually scripts/transcripts) that cater to busy professionals and curious minds alike. Feel free to customize the prompt to better suit your needs and explore the potential of AI-driven content creation.
An Example: A Podcast about Prompt Engineering
There are three ingredients to generate these scripts: the PROMPT (above), a source (a document, image, text file, PDF, etc.) to talk about, and an LLM to generate the response. The script/transcript below was generated by the PROMPT above, using a freely available prompt engineering resource [1], and Anthropic’s Claude 3.5 Sonnet. Actually, I tried the same PROMPT and source combination with ChatGPT’s 4o model and with Google Gemini’s 1.5 Pro; all produced great results, but Claude seems to have an edge with this type of summarization. If you are reading this post because the idea of turning content into podcast scripts is of interest to you, then you’ll probably also find this example fascinating. I promise that Mark and I don’t use this to generate our show. But perhaps we oughta.
The AI Insider (fictional podcast): Demystifying Prompt Engineering
Sarah (Explainer): Welcome to The AI Insider, your weekly dose of cutting-edge artificial intelligence insights. I'm Dr. Sarah McQueen.
Alex (Questioner): And I'm Alex Johnson. Today, we're diving into a topic that's been buzzing in the AI community: prompt engineering. Sarah, what exactly is prompt engineering, and why is it suddenly so important?
Sarah: Great question, Alex. Prompt engineering is essentially the art and science of crafting instructions for large language models, or LLMs, to get them to produce the output we want. It's become crucial as these models are increasingly integrated into various applications.
Alex: Interesting! So, it's like learning how to "speak AI," right? How does this process actually work?
Sarah: That's a great way to put it, Alex. At its core, prompt engineering involves four key components: instructions, context, input data, and an output indicator. Let's break these down.
Alex: Alright, walk us through it, Sarah.
Sarah: Sure thing. First, you have the instructions, which are clear directives telling the model what to do. Then there's context, providing relevant background information. Next, you have the input data, which is the actual content the model will work with. Finally, there's the output indicator, specifying what kind of response you're looking for.
Alex: Hmm, can you give us an example to illustrate this?
Sarah: Absolutely. Let's say we want to classify a restaurant review. Our prompt might look like this: "Classify the following review as positive, neutral, or negative: 'The food was delicious, but the service was slow.' Sentiment:"
Alex: Oh, I see. So the instruction is to classify the review, the context is that we're looking at restaurant feedback, the input data is the actual review, and the output indicator is the word "Sentiment:" at the end.
Sarah: Exactly! You've got it, Alex.
Alex: Now, I've heard that there are some settings you can adjust when working with these models. What can you tell us about that?
Sarah: You're right, there are two key parameters to consider: temperature and top-p. Temperature controls how random or creative the model's responses are. A lower temperature produces more focused, deterministic answers, while a higher temperature encourages more diverse and creative responses.
Alex: Fascinating! And what about top-p?
Sarah: Top-p, also known as nucleus sampling, narrows down the range of possible word choices. It helps control the breadth of the generated text, allowing for more or less variability in the output.
Alex: So, if I'm understanding correctly, you'd use different settings depending on whether you want a more factual or creative response?
Sarah: Precisely! For instance, if you're working on a creative writing task, you might set a higher temperature, like 0.7. But for fact-based responses, you'd want a lower temperature, maybe around 0.2.
Alex: That makes sense. Now, Sarah, I'm curious about the different tasks prompt engineering can be applied to. Can you give us some examples?
Sarah: Of course! Prompt engineering is incredibly versatile. It can be used for text summarization, question answering, text classification, and even code generation, among other tasks.
Alex: Wow, that's quite a range! Could you elaborate on one of those?
Sarah: Sure, let's take text summarization as an example. Say you have a lengthy medical article, and you want to get the key points quickly. You could craft a prompt asking the model to summarize the main ideas in one sentence. This could be incredibly useful for busy professionals who need to stay updated but don't have time to read entire articles.
Alex: That sounds really practical. Are there any advanced techniques in prompt engineering that our listeners should know about?
Sarah: Absolutely, Alex. Three techniques that have shown promising results are few-shot prompting, chain-of-thought prompting, and self-consistency.
Alex: Those sound intriguing. Can you break them down for us?
Sarah: Certainly. Few-shot prompting involves providing a few examples in the prompt to guide the model. It's like giving the AI some context to work with. Chain-of-thought prompting encourages the model to explain its reasoning step by step, which is particularly useful for complex problem-solving tasks.
Alex: And what about self-consistency?
Sarah: Self-consistency is a technique where the model generates multiple reasoning paths and then selects the most consistent answer. It's particularly effective for tasks involving common-sense reasoning.
Alex: These techniques sound powerful. Are there any tools available to help with prompt engineering?
Sarah: Yes, there are several libraries and integrated development environments (IDEs) that support prompt engineering. These tools allow users to experiment with different prompts, evaluate their effectiveness, and even manage prompt versions.
Alex: That's really helpful. Now, looking towards the future, what trends do you see emerging in prompt engineering?
Sarah: Great question, Alex. One exciting development is Reinforcement Learning from Human Feedback, or RLHF. This technique aims to align model responses more closely with human preferences. We're also seeing a growing focus on multimodal prompting, which involves working with models that can handle not just text, but also images and video.
Alex: Fascinating stuff, Sarah. But I have to ask - are there any concerns we should be aware of?
Sarah: Absolutely, Alex. As with any powerful technology, there are safety concerns. Two significant issues are prompt injection and jailbreaking, which can potentially manipulate model outputs or bypass safety protocols. These are active areas of research in the AI security community.
Alex: It sounds like prompt engineering is a complex and rapidly evolving field. Any final thoughts for our listeners?
Sarah: I'd say that prompt engineering is becoming an essential skill in the AI toolkit. It's not just about getting better results from models; it's about understanding how to communicate effectively with AI systems. As these models become more integrated into our daily lives and work, the ability to craft effective prompts will be invaluable.
Alex: Well, there you have it, folks. A deep dive into the world of prompt engineering. Thanks for tuning in to The AI Insider. I'm Alex Johnson.
Sarah: And I'm Dr. Sarah McQueen. Until next time, keep exploring the fascinating world of AI!
If you have a question you’d like to ask, or you made a helpful modification to the PROMPT that you’d like to share, or follow along with the discussion, please look for a thread about this post in the “Genealogy and Artificial Intelligence (AI)” group at Facebook. Thanks, Steve
Appearances
Co-host Mark Thompson and I will be Dana Leed’s guest as she livestreams from Baltimore on Friday 4 October 2024 for East Coast Genetic Genealogy Conference (ECGGC); catch her YouTube channel at 11 AM ET for about an hour of AI genealogy discussion and Q&A.
I will be speaking at ECGGC at 7 PM ET, Friday 4 October 2024 with Mark Thompson; this talk will be livestreamed for Conference registrants.
[1] Dr. Elvis Saravia < https://www.linkedin.com/in/omarsar/ > releases his Prompt Engineering Guide freely under an MIT license for re-use such as this exercise.
