WEBVTT

00:00:11.000 --> 00:00:14.000
this particular topic.

00:00:14.000 --> 00:00:32.000
Let's see, it is right about one o'clock. Just let me know where you're tuning in from, what you hope to learn from this session. The reason I'm so excited to talk about this is because I am a writer, and I love writing, and I'm really passionate about writing well

00:00:32.000 --> 00:00:42.000
So let me know where you're tuning in from. Say hello in the chat. Let me know. You can see me and hear my screen.

00:00:42.000 --> 00:00:46.000
Once I get something in there, I'll get started. Just want to

00:00:46.000 --> 00:00:51.000
I want to know you guys can see and hear me

00:00:51.000 --> 00:01:06.000
Awesome. Thank you, Sandra. I really appreciate that. Because we have so much to cover, while you let me know what you're hoping to take from this in the chat, I want to talk a little bit about why it is so important

00:01:06.000 --> 00:01:16.000
that you be able to maintain your unique tone and voice while you are using AI writing tools to help speed up the process. There are a couple different reasons

00:01:16.000 --> 00:01:31.000
The first reason that this is so important is because the way you sound absolutely influences the extent to which people trust you and the nonprofit that you work for. This is not a me on one survey to be very clear

00:01:31.000 --> 00:01:46.000
But in a survey recently launched, and I'll link to this in the follow-up, around a third of the people surveyed indicated that they would trust a brand less if they knew that its content was AI generated.

00:01:46.000 --> 00:02:01.000
And when you're a non-profit, you want to build trust with people, so that is not something that you want. Now that said, that was for any brand. The survey was not focused on nonprofits. It's probably most applicable for for-profit companies

00:02:01.000 --> 00:02:04.000
So what about nonprofits?

00:02:04.000 --> 00:02:22.000
Well, we decided to find out. We surveyed around 2,000 donors to ask them about their thoughts and opinions on AI in the nonprofit space specifically. Donors were most open to nonprofits being using AI for things like

00:02:22.000 --> 00:02:32.000
fraud detection and data analysis, and when it came to communications and other kinds of interactions, they expected those to be much more human

00:02:32.000 --> 00:02:49.000
Now, that doesn't mean that there is no room for you to use AI tools to help you write, and you're going to hear me emphasize the word help a lot in this content today, because if you're going to use AI for writing at all

00:02:49.000 --> 00:02:59.000
These findings mean that you need to take some steps to ensure that your tone, your voice, even the words and punctuation that you use are yours

00:02:59.000 --> 00:03:14.000
Donors want to hear from you and from your organization. They do not want to hear from ChatGPT. I love that Maria called out that they are hoping not to have their words scored at 90 plus percent on Pangram

00:03:14.000 --> 00:03:32.000
That's fantastic. Okay, let's see. Why isn't my slide progressing? That's cool. All right, so people want to hear from you, but in addition to that desired or that need to meet people's desire for human interactions

00:03:32.000 --> 00:03:38.000
There are some concerns that you want to keep in mind if you choose to use AI to help you write.

00:03:38.000 --> 00:03:55.000
Now, to be clear, I do use AI sometimes, but I am first and foremost a writer. And so I feel comfortable telling you that AI as a technology can support you in the writing process, but it should absolutely never, ever, ever

00:03:55.000 --> 00:04:03.000
replace you. I don't want to get very deep in the weeds of how LLMs work

00:04:03.000 --> 00:04:20.000
But you can think of an LLM like what you would use for generating content in ChatGPT or Claude or something similar as very sophisticated autocomplete on your phone. LLMs are trained on huge, huge bodies of writing, and then they use an algorithm

00:04:20.000 --> 00:04:28.000
to predict what word is most likely to come next in a sentence based on the prompt you've given it.

00:04:28.000 --> 00:04:35.000
So, being trained on all of that data means that plagiarism is going to be something that you'll need to keep in mind

00:04:35.000 --> 00:04:47.000
plagiarism is a big issue with LLMs. Even if you don't intend to plagiarize someone, it is easy for that to happen simply because of the way those LLMs are built and trained.

00:04:47.000 --> 00:05:03.000
Large language models like ChatGPT and Claude can also have a hard time with specificity, and it can also have a really hard time with things like numbers and statistics and citable facts.

00:05:03.000 --> 00:05:19.000
Because of this, it is possible for LLMs to hallucinate a lot. I think hallucinate is a very fun word to describe what's happening. But LLMs don't know things, despite the phrase artificial intelligence, this technology is not intelligent. It cannot think or reason

00:05:19.000 --> 00:05:27.000
LLMs don't know things, they only predict what is most likely to come next in a sequence of words.

00:05:27.000 --> 00:05:47.000
So if you are using an untrained LLM, and we're going to get into that in a minute, and ask it to help you write an impact report, there is a very real chance that it will make up numbers or make up impacts or make up stories. That's something to be concerned about, especially as you are trying to build trust with your community.

00:05:47.000 --> 00:06:06.000
You may also encounter the phenomenon that I've seen some people refer to as beige goo. If you are familiar with beige goo, I would love to hear about it. It is one of the funnier phrases I've heard around this. Beige goo is a phrase that I have also adopted. I did not originate it, but I have adopted it

00:06:06.000 --> 00:06:15.000
From other people and it really just described the homogeneity or the sameness of AI generated content.

00:06:15.000 --> 00:06:28.000
LLMs will frequently use the same words over and over and over again, even if those words aren't something that real human writers would use.

00:06:28.000 --> 00:06:44.000
LLMs also tend to use very repetitive sentence structures. If you are a word nerd like me, one, let me know in the chat if you are a fellow word nerd, because I would love to talk with you. The syntax, the order in which it puts words is

00:06:44.000 --> 00:07:01.000
stilted and can be very awkward. These models tend to reuse the same punctuation, even if that punctuation is not the most appropriate for the use case. And people can tell. They may not be able to point to a sentence and say, this use

00:07:01.000 --> 00:07:16.000
of a present participle after an independent clause is a sign that this is an AI generated piece of content, but people do absolutely pick up on the fact that content is generated by a machine and not by a human person.

00:07:16.000 --> 00:07:33.000
This quote kind of sums it up. This is a quote from a paper. It's got a very catchy title. It's called Differentiating Between Human Written and AI-generated Text Using Automatically Extracted Linguistic Features. This was published last year. The title does not exactly trip off the tongue, but it is very informative

00:07:33.000 --> 00:07:52.000
And the quote just says, AI language does not follow the exact patterns found in human language. There are significant linguistic distinctions between human written and AI generated content. All of that is to say that someone may not be able to explain why a piece of content

00:07:52.000 --> 00:07:53.000
feels off

00:07:53.000 --> 00:08:13.000
But they will know that it feels off, and that can have a negative impact on the way they read it, and the way they perceive what you have to say. So it is important, as you use these tools to enhance the writing process, that you, you remain the center of anything

00:08:13.000 --> 00:08:34.000
you create. So how do you sound like a human person when you are using a robot to help you write? And I know that it's not actually a robot. I know that it's an algorithm, but we'd like to personify things here. The first step is to train the tool you're using, Cynthia. This is for you because you have this beautiful question

00:08:34.000 --> 00:08:49.000
The first step is to train the tool you're using and full disclosure, the tool that you're using, and in some cases, whether or not you are using a paid version of that tool can impact the ability to which you can do this.

00:08:49.000 --> 00:08:58.000
So, training your model, and I want to make this clear before I tell you how to train a model, training your model is not going to make it perfect.

00:08:58.000 --> 00:09:06.000
You will still need to edit anything a model gives you, even if you have trained it

00:09:06.000 --> 00:09:21.000
But a well-trained model will give you a first draft that is closer to what you want. It will sound less beige gooey. There will still certainly be some some tells that something is AI generated. We'll go over those tells in a few minutes

00:09:21.000 --> 00:09:37.000
But this will sound more like your organization, and you will ideally spend less time on the rewriting and editing portion of the process if you take this step. So what does that look like? I'm going to tell you what I have done

00:09:37.000 --> 00:09:54.000
I have done this myself. I've created something that I refer to as Abby GPT. One of our other writers here at Neon One did the same thing, but he named his model Wario. So whether or not you decide to name yours, you're going to follow some of the same steps

00:09:54.000 --> 00:09:57.000
The first thing you're going to do is choose your tool

00:09:57.000 --> 00:10:14.000
if you are using AI for writing, please let me know what tool you're using. I'm actually quite curious. I prefer Claude. I also have used Gemini. I've also used ChatGPT. I prefer Claude. So, choose your tool

00:10:14.000 --> 00:10:18.000
And then tell your tool what it's for

00:10:18.000 --> 00:10:34.000
And the more specific here you can be, the better. So my tool, for example, is I have a project in Claude. My project is only used to repurpose original content that I have created myself

00:10:34.000 --> 00:10:50.000
So I told it that I write for nonprofit professionals like you. I tell it what I want it to sound like. I've given it rules that it should follow. I have given it hard guidelines around what it may and may not do

00:10:50.000 --> 00:10:56.000
You're going to want to do something similar. I see that we have a pretty big mix of folks here

00:10:56.000 --> 00:11:05.000
I wish I preferred Claude, but ChatGPT is my tool. Claude, I found ExtraWordy Kelsey. Claude can absolutely be very wordy.

00:11:05.000 --> 00:11:20.000
So you're going to want to do something similar. You're going to want to tell your tool who your audience is, what your organization is about. Maybe you want to share your mission and vision statements, and you're going to want to create a series of rules and guidelines to follow.

00:11:20.000 --> 00:11:30.000
Here is a tip that I did not include on this slide. If you are curious about what you need to give your tool, try asking it

00:11:30.000 --> 00:11:41.000
Tell her that you are… you want to train it to do the specific task, what information does it need from you to do a good job and to get your first drafts closer to a finished product?

00:11:41.000 --> 00:11:53.000
Once you have described your tool to itself and trained it a little bit here, you're going to want to give your tool reference materials. And Cynthia, this is what I think you were getting at.

00:11:53.000 --> 00:12:09.000
I would recommend giving your tool emails, letters, and content that your organization sends to your audience. I wouldn't give it anything that you send kind of internally, unless you're using that tool exclusively for internal communications.

00:12:09.000 --> 00:12:28.000
Website content, also very helpful to give to your tool. One of the references I gave my own tool when I was building Abbey GPT was a several hundred page document of all of the blog articles that I had written for Neom 1.

00:12:28.000 --> 00:12:40.000
All of those blog articles were written by me without AI help. And then I asked it to make observations about my writing style and content, and then I used that feedback to train the model again.

00:12:40.000 --> 00:12:55.000
And don't worry, I'm going to show you what this looks like. If you have any educational content that you reference in grants, if you have any, don't put, like, personal information in there or information about your grant funders. I only mean the educational content

00:12:55.000 --> 00:13:05.000
Or if you have impact reports that you also have some statistics that you may want to pull from, that is also really valuable. Again.

00:13:05.000 --> 00:13:23.000
I'm telling you to give your model references to help train it, but you are always going to want to fact check every single piece of content you create with this tool's help. Giving it reference material will help prevent hallucinations, but it doesn't entirely prevent them from happening.

00:13:23.000 --> 00:13:39.000
I love this note from Mara. I have it asked me for other information it needs to start. It will usually ask me about audience, format, tone, etc. It knows I'm fundraising, so it will ask about donor preference. That's extremely helpful.

00:13:39.000 --> 00:13:47.000
This is an example of what that might look like for a little while. We were using Gemini. We've since moved to Claude for content creation, but this is what I did

00:13:47.000 --> 00:14:02.000
So I named Abby GPT, I gave it extensive instructions. This is in Gemini, it's phrased differently. Who you are, I want you to be my digital twin. Here's a little bit about me. I told it how long I'd been in the tech sector. I told it about who I talked to

00:14:02.000 --> 00:14:12.000
I gave it grammatical rules. Again, I'm a word nerd, I have a literature degree, I have really strong opinions about punctuation and phrasing. There are a lot of rules here

00:14:12.000 --> 00:14:30.000
And then in knowledge here, and you'll see kind of a similar area in Claudia and ChatGPT. I gave it a couple pieces of research that I had written myself, and then I gave it this file, which is like a 500 page download of all of the copy I had created

00:14:30.000 --> 00:14:34.000
On our blog. And I use this robot

00:14:34.000 --> 00:14:40.000
specifically for content creation for my job. So, all of this is like

00:14:40.000 --> 00:14:46.000
intended to train the model to do the very specific tasks I am building it to do.

00:14:46.000 --> 00:15:01.000
All right, so you have your tool, it knows who you are, it knows who you're talking to, it knows what it's for. It even knows a little bit about your writing style, your organization, your preferences. But even the best trained tool can get you only so far

00:15:01.000 --> 00:15:19.000
If you want to create content using an AI tool that is closer to a finished draft, never a finished draft, and not a very generic one, you're going to want to take your time with your prompts, and you're going to want to write prompts that are very thorough.

00:15:19.000 --> 00:15:34.000
You can use AI tools for all kinds of things. Donors do not want final drafts that are AI generated, but they are generally pretty open to fundraisers using AI for things like analysis or brainstorming

00:15:34.000 --> 00:15:50.000
They are pre-opened about, like, outlining content. I personally get the most use from AI as an editor. I generally prefer to write things myself, but I will use an AI tool to do an initial edit

00:15:50.000 --> 00:15:57.000
It's really great at catching typos and little grammatical mistakes. But regardless of what you're doing, the more

00:15:57.000 --> 00:16:02.000
you are in your prompt, the better the results will be at the end.

00:16:02.000 --> 00:16:25.000
So here, don't panic at all the tests on this, on this page. Here are a few pointers for writing good prompts. The first thing to do is to be very specific. You may have heard that it is most useful to treat AI tools like an untrained intern, a very enthusiastic intern, but an untrained intern nonetheless. And honestly, I think that's really great advice.

00:16:25.000 --> 00:16:31.000
If you go to a brand new intern, a little

00:16:31.000 --> 00:16:47.000
A little before Giving Tuesday, and you say, hey, I want you to write, a blog article about what we achieved on Giving Tuesday and what that money is going to be used for. And that's the extent of the information you give your intern. You're not going to get a great result

00:16:47.000 --> 00:17:03.000
But if you go to your intern and you say, hey, I want you to write a blog article that updates our donors on this campaign that we ran, here's what the campaign was focused on, here are the primary talking points, here's what our fundraising goal was, here's how much we've raised, here is

00:17:03.000 --> 00:17:19.000
Approximately the structure I want you to use, or I want you to make sure you touch on these points, you're going to get something that's much closer to a finished product. This example here on the right is a prompt I recently wrote using Claude

00:17:19.000 --> 00:17:34.000
And I asked my LLM for a blog article. Now, because I am opinionated about this kind of thing, I knew exactly how I wanted the article to be structured. And I knew approximately what information I wanted it to include

00:17:34.000 --> 00:17:50.000
So I gave this tool an extensive outline. The reason I did this is because I have learned that Claude, while great at following detailed instructions, is not particularly adept at putting together kind of a linear structure for content

00:17:50.000 --> 00:17:58.000
So, I don't let it decide how it's going to structure my content. I tell it how to structure the content, and then it gives me more or less what I'm looking for.

00:17:58.000 --> 00:18:04.000
If you can, in your prompt, give it specific reference material

00:18:04.000 --> 00:18:16.000
So in this example, I gave my LLM directions to reference a copy of a research report I had already uploaded. So you can see it up here. Please pull from the resources I've uploaded to your context

00:18:16.000 --> 00:18:32.000
I do also say please to the robots. I think it's nice. This is important for a couple of reasons. One, giving it extensive reference material will help you prevent a lot of plagiarism and it will help you prevent hallucinations

00:18:32.000 --> 00:18:40.000
Especially if you are sharing concrete facts, statistics, or even very specific stories about your programs and your activities.

00:18:40.000 --> 00:18:48.000
Prompts are going to be most effective for you when you write them in short, clear, easy-to-understand sentences.

00:18:48.000 --> 00:19:04.000
I will also tell you it is okay, and even a good thing for you to treat your first prompt as one of a series. I will say that I will frequently create new chats for different parts of a drafting process

00:19:04.000 --> 00:19:16.000
Here, I asked Claude to give me an article based on this outline. I rewrote it, I started a new chat, gave it my article, and asked it to do a quick scan for typos and mistakes.

00:19:16.000 --> 00:19:24.000
when I have learned that when I do that, I get results that I like better, so I would encourage you to experiment with that

00:19:24.000 --> 00:19:33.000
Here are some general best practices as you are writing these prompts. One, remember, LLMs are fancy autocorrect.

00:19:33.000 --> 00:19:50.000
That sounds dismissive, and I don't mean it to be, but they are. They are writing algorithmically, they're not thinking through writing for you. Despite the name, the tools are not actually intelligent, so what you can do with your human brain, what you are good at, which is breaking down the process into steps

00:19:50.000 --> 00:20:06.000
That will get you better results than if you trusted it to think for itself, because it can't do that. Again, triple check everything it gives you, including statistics and facts and the commentary it adds around those

00:20:06.000 --> 00:20:20.000
So in this conversation here on the right, this is exactly why this is important. I gave Claude a file and I asked it to recommend 15 blog topics. I just wanted some brainstorming help. In three cases, Claude either made mistakes

00:20:20.000 --> 00:20:28.000
with statistics, it just… it gave me the wrong number, or it misinterpreted what the statistics meant.

00:20:28.000 --> 00:20:37.000
So even with a good prompt and very solid reference material, your first draft will probably contain mistakes.

00:20:37.000 --> 00:20:51.000
As you get comfortable with your tools, experiment with prompt structures. I realized I have very strong preferences about how my content is structured, so I don't ask Claude to do that for me. I do it. And then

00:20:51.000 --> 00:21:09.000
It kind of builds upon that scaffolding that I've given it. Another thing I would encourage you to do, this is actually top of mind for me this week, remember that even though some tools can remember past conversations, it's best to treat them as though they have terrible memories.

00:21:09.000 --> 00:21:17.000
I get the best results from my tools when I start new conversations from scratch instead of picking up old threads

00:21:17.000 --> 00:21:28.000
And you can experiment with that too. Here is the most important part of the whole process. I've been alluding to this the whole time. It is absolutely critical that you rewrite and edit anything that you get

00:21:28.000 --> 00:21:43.000
Because you could give your tool the most gorgeously written prompt of all time. You could give it to the most well-trained model around, and your output will still sound like it was written by AI, because it was written by AI

00:21:43.000 --> 00:21:51.000
And that's not bad. That is the tool functioning as it is supposed to. You are the human writer, and you are just using it as a tool.

00:21:51.000 --> 00:21:59.000
The chat your GPT or whatever you're using will still hallucinate or misunderstand what you've given it.

00:21:59.000 --> 00:22:14.000
For example, if you guys read, the generosity report that we published last year, I was using my tool to help me write some content around it, and it said that the report included step-by-step strategies for donor retention, which it did not. So

00:22:14.000 --> 00:22:21.000
these… remember, these tools are fallible. They will make mistakes. It's our job to catch those mistakes

00:22:21.000 --> 00:22:38.000
So, here's my favorite part, because I want you to talk to me in the chat. As you rewrite and edit, keep an eye out for some telltale characteristics of AI writing, and I am so curious what you look for because I am about to give you a lot of examples of things that I look for in my own writing

00:22:38.000 --> 00:22:44.000
What do you notice in AI generated content that you will always change

00:22:44.000 --> 00:23:01.000
Here are some common tells that written content is generated by AI. Repetitive sentence structure. I don't know how many times you see in this example here, it's the not only, but also it's not about blank, it's about blank

00:23:01.000 --> 00:23:12.000
LLMs use this sentence structure all the time. They also love the rule of three, which is where you will cite three examples or include three items in a series

00:23:12.000 --> 00:23:32.000
Oh my gosh, Danielle, unwavering commitment. So many people have the MDASH. M-dash makes me sad because I love the MDASH. I read a lot of Emily Dickinson when I was in middle school, and it completely changed the way I write. Humorously, LLMs use MDASH in a very specific and always incorrect way, which I think is very funny

00:23:32.000 --> 00:23:45.000
LLMs tend to love using present participle. So Neon CRM is a great platform, comma making it the perfect tool for you. That is not something you frequently hear in human speech.

00:23:45.000 --> 00:24:00.000
I love everybody who's revolting, like, we're keeping the M dash, it's ours. LLMs frequently use very short, repetitive sentences or sentence fragments in an attempt to elicit an emotional response

00:24:00.000 --> 00:24:11.000
They will frequently use very weirdly friendly language that I would never use. Oh yes, Shannon, the meeting of the MDASH fan club is officially called to order

00:24:11.000 --> 00:24:29.000
We will have decoder rings and t-shirts. You will notice the first bullet point on this one is misused em dashes. LLMs frequently misuse them, which is how I can often tell that something is AI generated and not something written by a human who loves this extremely useful

00:24:29.000 --> 00:24:30.000
punctuation mark

00:24:30.000 --> 00:24:34.000
It is frequently

00:24:34.000 --> 00:24:50.000
irregular when it comes to using other types of punctuation. I will notice, for example, I always use an Oxford comma, but I will notice in drafts given to me by Claude that sometimes it will use an Oxford comma, sometimes it will not. It frequently mixes up colons and semicolons

00:24:50.000 --> 00:25:10.000
things like that. I love the note that there's overly casual content that drives me bananas. I frequently also notice very strange formatting. For whatever reason, Claude is on a real tear right now, or it will just bold random sentences for me, which I dislike, and I've told it a couple of times to stop doing that.

00:25:10.000 --> 00:25:29.000
One thing that I have also noticed is models will frequently choose a word and overuse it for a long time. Right now, everything, I see quietly everywhere. I have never, ever used quietly to describe something other than the level of noise associated with an activity.

00:25:29.000 --> 00:25:50.000
I love how many people are calling out different things. Beacon of Hope is a real big one. There are repetitive words. And the thing that I love most about this is that while there are a lot of us who are noticing the same things, a lot of us are frequently encountering or noticing different trends and patterns

00:25:50.000 --> 00:26:03.000
This is because LLMs are always learning, and their tricks and their tells will change. They used to use, for example, emojis just willy-nilly all over the place. That has since been less common

00:26:03.000 --> 00:26:21.000
Now they're using words like quietly or softly or something like that. As you get better at using your AI tools, you will start to notice those ticks and trends changing. You will notice them fading away. You will notice new ones. Here is my number one tip for you

00:26:21.000 --> 00:26:26.000
As you are writing or drafting, if you get stuck

00:26:26.000 --> 00:26:45.000
Read your work aloud, and if that's not how it would sound when you speak it out loud, if you wouldn't say those words out loud to the donors or supporters or whomever you're talking to, just change it. The I sound very silly on a regular basis, because I will read things aloud to make sure that they sound correct

00:26:45.000 --> 00:26:47.000
Once you have your final draft

00:26:47.000 --> 00:27:03.000
review it for structure. Again, a lot of LLMs really struggle with structuring good content. If you have an appeal, you know how it's supposed to go. You have the story or the hook that grabs someone's attention, you tell them about the need, you make a case for support

00:27:03.000 --> 00:27:19.000
And then you close it out. That is going to be mixed up a lot. So review it for structure and reorganize it as you need. Remove or remediate a lot of the tells that we have discussed. Switch up sentence structure

00:27:19.000 --> 00:27:34.000
for goodness sake, if you can, avoid the juxtapositional, it's not about blank, it's about blank, language. That's a huge tell that people pick up on. Bad punctuation, inappropriate formatting, the weirdly informal or casual content that a lot of you called out

00:27:34.000 --> 00:27:45.000
Anything you get is a rough draft that needs extensive work. You are never going to get a final product out of an AI-generated tool. It can help you speed up the process.

00:27:45.000 --> 00:28:00.000
starting to write from a blank page is really hard, and sometimes something like this can help get you started. I know it makes it a lot easier for me to create content. I can create content much more quickly, but I still have to spend time making that content

00:28:00.000 --> 00:28:08.000
good and useful and human sounding instead of it being ChatGPT or Claude or Gemini.

00:28:08.000 --> 00:28:24.000
This is an optional step. This is most useful if you are using a paid version of any of these tools that you can use, where you can train it more extensively. Update your model. So once you've edited and refined your content

00:28:24.000 --> 00:28:38.000
give it back to the AI tool that you're using and ask it to make observations about the changes and updates that you've made, and then use your feedback to go back and update the instructions. So here

00:28:38.000 --> 00:28:53.000
For example, this is Claude. I told it like, please pull it. It noted that I changed percentages to be exact. I like exact percentages. Instead of just talking about relative percentages. I

00:28:53.000 --> 00:29:10.000
We'll all in this in particular, I will frequently affirm a reader's overwhelm when talking about this particular topic. And just open with an affirmation that they're probably feeling overwhelmed, reassured them that we can do it together and then go into the rest of my content

00:29:10.000 --> 00:29:27.000
And then I reframed it, reframed it again. At the bottom, like, we are positioning this advice as something that is supposed to be useful long-term, is not super urgent, and then I took all of its observations, took it back to the instructions that I had given my tool

00:29:27.000 --> 00:29:41.000
And then updated the instructions that way. And as you use and refine your tool more and more over time, again, you will get even closer to a finished product every time. Even though, again, that's never a complete product, it's always a rough draft.

00:29:41.000 --> 00:29:56.000
All right, do you have any questions for me? That was a lot of information in 29 minutes. I love this. I move a document from ChatGPT into Word for editing. The AI in Word wants to make lots of grammatical changes

00:29:56.000 --> 00:30:05.000
That is always kind of a hoot, because you do see different grammatical styles in different models.

00:30:05.000 --> 00:30:20.000
I like Claude most, because it most closely aligns with my style preferences. I have noticed that, like, ChatGPT tends to format things much more as you would in, like, an AP or in science writing, which makes sense

00:30:20.000 --> 00:30:29.000
Claude seems much more oriented towards humanities, so I get that. How do you give ChatGPT reference material like emails or annual reports

00:30:29.000 --> 00:30:32.000
I

00:30:32.000 --> 00:30:35.000
Hold on, I'm going to

00:30:35.000 --> 00:30:39.000
I'm gonna look up something here.

00:30:39.000 --> 00:30:55.000
What I'm doing is I'm trying to go to OpenAI right now. The challenge is that I do not have a paid OpenAI subscription, nor do we allow OpenAI usage at Neon 1. So I cannot actually pull up OpenAI

00:30:55.000 --> 00:31:05.000
Right now, and tell you how to do that. But what you can do in almost any tool you use, if you are, especially, I think you may have to use a paid version of these tools

00:31:05.000 --> 00:31:24.000
There is the ability for you to create, I believe they're created… I think they may be called skills in ChatGPT. If someone is a ChatGPT user, please remind me what that is actually called. But you will be able to give it a reference materials, and in that kind of builder, you should be able to upload files

00:31:24.000 --> 00:31:34.000
That way, if someone is a ChatGPT user and can give Kennedy more specific advice there, I would love that. But again, because we don't use OpenAI

00:31:34.000 --> 00:31:42.000
Pat me on one, I cannot do that for you right now. I can show you in Claude what you can do

00:31:42.000 --> 00:31:56.000
Is you'll go to Projects, you'll start a new project. You tell it what it is, what you're working on, what you're trying to achieve. You'll create your project, and then you can upload files there.

00:31:56.000 --> 00:32:10.000
It is something… let's see, let's go on a journey together. Let's… because I see that, Kelsey is having the same question. Train a chat

00:32:10.000 --> 00:32:12.000
GPT.

00:32:12.000 --> 00:32:18.000
If you go here, well, that's not particularly useful.

00:32:18.000 --> 00:32:36.000
So you can go into your profile settings. I am reading the AI generated summary, because I can't access ChatGPT on my computer. You can go to your ChatGPT, build out your profile, and kind of include instructions that way. I believe that is also where you would do that.

00:32:36.000 --> 00:32:40.000
You can also build

00:32:40.000 --> 00:33:00.000
Let's see, I found a very useful Reddit thread. You can look at this together. You can build a knowledge base in a custom GPT. Oh, thank you, Marath, and thank you, Cynthia. Mara says there are projects in ChatGPT. That's also what they're called in Claude. And then Cynthia said that ChatGPT has a library where you can upload documents

00:33:00.000 --> 00:33:20.000
As you are learning to do that, I do know that OpenAI has some pretty solid documentation, and there's also a really active ChatGPT subreddit where you can probably find some basics. So that's the tricky part of learning ChatGPT across multiple systems. I have used ChatGPT in the past

00:33:20.000 --> 00:33:39.000
But it has changed so much in the year or so that I haven't used it that I'm not familiar with it anymore. So thank you all for that. I really appreciate it. Okay, we have reached the end of our questions, so here's what's going to happen next. The perpetual question I get at every session is, are you going to send me a recording? And the answer is yes

00:33:39.000 --> 00:33:57.000
We have been recording this. This is going to show up in your inbox tomorrow morning. You'll get an email from me with a link to this. You'll get a link to the slides, and then we've included some additional resources that you can use as you use AI to enhance the writing process

00:33:57.000 --> 00:34:12.000
So keep an eye out for me for that. And then I think that's everything. I hope you all have a wonderful Wednesday. I hope you are getting excited about the end of the year and I hope you feel a little better equipped

00:34:12.000 --> 00:34:31.000
To use AI tools to help you write smoothly, but in a way that maintains the beautiful human perspective that you bring to your writing. I appreciate you all so much. I will talk to you tomorrow morning. Keep an eye out for me. Have a great Wednesday.

