March 4, 2025 • 25 mins
AI is transforming the way we work—but how can Thinkydoers use generative AI responsively and productively—as a tool, not a time-suck?
In part two of our conversation with AI ethicist and strategist Dvorah Graeser, we get practical about leveraging AI tools, particularly for deep specialists and strategic thinkers. Building on our discussion of ethics and implications from Episode 38, Dvorah shares specific ways to move from anxiety to agency with AI while maintaining your ethics and humanity. Learn how to use AI tools to organize your thoughts, improve communication with different audiences, and enhance your work as a specialist, all while preserving what makes us uniquely human.
Episode Highlights:
- How AI can help specialists move from anxiety to agency and practical ways to embrace AI tools with confidence
- Using AI to organize complex ideas and turn scattered thoughts into structured outlines
- The power of AI iteration and dialogue to refine thinking, decision-making, and communication
- Fact-checking and avoiding AI hallucinations with tools like Perplexity AI and multi-tool validation
- Enhancing communication for deep specialists by adapting messaging for different audiences, including neurodivergent thinkers and general business professionals
- Leveraging AI for buyer personas and using marketing strategies to improve AI-generated content and audience targeting
- AI as an assistive tool for neurodivergence and how it supports executive functioning challenges and workflow efficiency
- Emotional processing and AI as a tool for self-reflection while recognizing its limitations in mental health support
- First steps to overcoming AI anxiety and why understanding your emotional response to AI is crucial before integrating it into your workflow
- Resources and further learning, including Dvorah’s free PDF on emotions in AI and an invitation to submit questions for future discussions
Key Concepts Explored:
AI for Organization & Creativity
- How AI helps specialists structure ideas and overcome creative overwhelm
- The role of AI in iterative thinking and refining communication
- AI’s effectiveness in summarizing, outlining, and organizing complex information
AI Accuracy & Fact-Checking
- Strategies for verifying AI-generated content and reducing hallucinations
- Using multi-tool validation with AI models like Perplexity AI and Claude
- Limitations of AI in handling data accuracy and research citations
AI for Specialists & Communication
- How AI helps deep specialists communicate with broader audiences
- Leveraging buyer personas to tailor AI-generated messaging
- Using AI to bridge communication gaps between neurodivergent and neurotypical audiences
AI for Neurodivergent Thinkers
- AI as an adaptive tool for executive function and workflow management
- The role of AI in supporting focus, organization, and stress reduction
- Ethical considerations of AI in self-reflection and emotional regulation
Overcoming AI Anxiety & Ethical Adoption
- The shift from AI anxiety to AI agency and practical steps for confident adoption
- The importance of emotional awareness in integrating AI into personal and professional workflows
- Resources for understanding the emotional impact of AI and strategies for responsible use
Episode Chapters:
00:00 Introduction to Thinkydoers Podcast and Episode Overview
01:00 From AI Anxiety to AI Agency – How Specialists Can Take Control
02:00 Organizing Chaotic Thoughts with AI – Turning Sticky Notes into Structure
04:00 The Power of Iteration – Using AI for Back-and-Forth Thinking
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:01):
Welcome to the Thinkydoers podcast.
Thinkydoers are those of us drawn todeep work, where thinking is working.
But we don't stop there.
We're compelled to move the work frominsight to idea, through the messy
middle, to find courage and confidenceto put our thoughts into action.
I'm your host, Sara Lobkovich.
(00:23):
I'm a strategy coach, a huge goal settingand attainment nerd, and board-certified
health and wellness coach, working atthe overlap of work, life well-being.
I'm also a Thinkydoer.
I'm here to help others find moresatisfaction, less frustration, less
friction, and more flow in our work.
(00:46):
My mission is to help changemakers likeyou transform our workplaces and world.
So let's get started.
Welcome back to part two of myconversation with Dvorah Graeser.
In our last episode, weexplored some of the ethics and
implications of AI at work today.
(01:07):
Now, we're getting practical.
We'll discuss some specific and maybesurprising ways to leverage AI tools,
particularly for deep specialistsand those of us wired strategically.
If you're ready to move fromanxiety to agency with AI while
maintaining your ethics andhumanity, this episode's for you.
(01:33):
Dvorah, you mentioned kind of the shiftfrom anxiety about AI to agency and
how people can leverage AI around theircreative vision, regardless of what
their technical level of proficiency is.
Tell me a bit more about howyou see the use of AI enabling
(01:53):
creative and multi-talented humans.
Well, one way that I personally like touse AI—I don't know if you've gotten to
this point when creating, written content,but I'll have a whole pile of thoughts,
and they are all over the darn place.
If they were sticky notes, they wouldbe like Mount Everest of sticky notes.
And I don't know what the heck to do withthem, and I get very frustrated by this.
One of the things that I have foundthat can be great to do is I shove the
(02:17):
whole thing into, let's say, Claude.
Now, Claude has a feature where you cancopy and paste, but you can also upload.
So, let's say you have a PDF andyou have some written documents
and some other stuff, and you justdump the whole thing into Claude.
And I just say Claude exactly what I'mfeeling — I don't know how to organize
this, like, if I can't make sense of this,or I have one thought, I will write the
(02:40):
thought, but the rest of it's not clear.
What do I do?
And then Claude very calmly chugsalong and proposes something.
They propose an outline.
It can be helpful if you ask itfor something specific, like an
outline or whatever, but even ifyou don't, it'll still do its best.
And then you start a dialogue.
And this is one of the things Ireally do like about generative AI.
(03:00):
We have a dialogue, we go back and forth.
I ask for constructive feedback becauseotherwise, it'll only say nice things.
It's something about the generative AI.
If you ask for constructive feedback,it'll also say the mean things.
But I'm going back and forth, andas I'm going back and forth, in my
own mind, I'm realizing, I don'tlike that bit, I do like this bit.
Oh, I just thought ofsomething completely different.
(03:20):
Now, what went from a blizzard ofideas I couldn't organize will go to an
organization that maybe isn't exactlywhat I want, but helps clear my mind,
reduce my own anxiety, because it'soften my anxiety that's leading to the
pile-o'-sticky-notes problem, at least inmy experience — and enables me to get to
the point where I can do the organization.
Even if it turns out that it'snot Claude that has made the
(03:42):
good proposal by interacting withit, I get to where I need to go.
That is the use case that has beenthe most transformative for me.
And that's where I went from being askeptic who was not likely to integrate
these tools to being a hooked user.
They don't replace the human brain.
(04:02):
You mentioned hallucinationsin our first episode.
I don't rely on models for facts orinformation, but they are really good
at organizing big messes of information.
The thing that frustrates me is sometimesthey're poor at actually organizing data.
(04:23):
Even when I train it on data andwant data analysis or data, you know,
stuff, that's been a big frustrationfor me with the current tools.
Even when I train on data, it canstill hallucinate, which is a bummer.
That's the kind of stuff that Iwould think the models would be
built to do really well, and withas little hallucination as possible.
(04:46):
But maybe there's always goingto be some hallucination.
There's a couple of waysaround hallucination, right?
And this again gets into process.
We talked about it in the first, episode,but it's still important to bring up.
You need to have a processfor dealing with this.
So let us suppose you uploadeda bunch of data to, I don't
know, let's say it was ChatGPT.
It doesn't matter.
You uploaded it, andit's still hallucinated.
(05:07):
So what are ways to check this?
One way to check this is to ask ChatGPT inthe same session to go back, and to check
its answers in comparison to the dataand show where it got each answer from.
This is also great forsummarization, by the way.
If it summarizes and it comes upwith something, you say, "Okay,
where did that really wonderful,great quote, where is it?
(05:28):
Show me where it is." And they'relike, "oh yeah, you're right, I
kind of made that up." Okay, noproblems, let's just move on.
A third way for doing it is thento actually go to another tool like
Perplexity and ask it to do research.
So maybe you don't want to uploadyour sensitive data to Perplexity,
but you can still ask it to doresearch on the area and give you some
(05:50):
ideas of what the boundaries of thearea that can help corral ChatGPT.
Or you can go to acompletely different tool.
I guess the completely differenttool is the third way of doing it.
Let's say you go from ChatGPT to Claude.
You can go to Claude, and then you cansay, "Hey, can you please check this?
And what is good and what is bad, andplease give me constructive feedback."
That's wild.
In my early experimentation, when Iwas working on my book, there was some
(06:12):
research I had, like actual research.
I needed academic journal research.
that was one of the thingsthat was the most frustrating.
I would think that models would begood at knowing that truthful data.
Hopefully, this has changed since then.
But I would ask for research, like,"What are the leading papers or
(06:33):
core papers on a specific topic?"And it would make up papers.
You've got ProQuest.
It's all there.
This you should get right.
So, for example, Perplexity AI,I like because it gives links.
And you can go check them.
ChatGPT sometimes does, sometimesdoesn't, sometimes hallucinates,
sometimes it's accurate.
I'm never quite sure what it's going to donext, so I find that a little unreliable.
(06:55):
There are some specializedGPTs for papers.
There's one called ScholarAI, and that's a good one, but
there are other ones as well.
So here, I think it's good to joinwith others to understand what they've
done and what's worked for them.
And that can change over time.
So, for example, I like Scholar AI,but it's possible in a few months,
someone's going to come back.
Maybe they'll listen to this again.
They'll be like, "Oh, but it'sno longer good." Possible.
(07:17):
Things change really quickly.
So I think it's helpful to try outdifferent specialty tools, to talk with
other people about this on LinkedInor another social media that you like.
Mastodon, if you're into that.
And try to get information about what isworking right now and what is the process
for working around some of these issues.
So if I get anything from ChatGPT,because I have found it to be
(07:38):
the most erratic when it comes tohallucinating, I will copy and paste
the whole answer to Perplexity AI.
And I'll say, "Is this right or wrong?
Please give me feedback.
Please find me references.
What do you think?" And it will gothrough the whole thing and it'll say,
"Well, this looks like it's correct.
Couldn't find anything for this.
I really don't know aboutthat bit over there.
And this is what I would do instead."And so Perplex AI will actually
(07:58):
give you a long, reasoned answer.
This was without going into theirdeep research mode, which is awesome.
But that's like a wholenother kettle of fish.
It's just regular Perplexity AI.
It'll actually go through these things.
So I guess it's unlike Google in thesense where if you get a link and you
click on the link, the link exists.
If you get a summary with thelink, does the link exist?
Does it not exist?
Does the summary say what the link says?
(08:20):
Not always.
So here we need to have more of aprocess for checking for these issues.
It can still, however, save a lot oftime because if you have a multi-tool
process and you leverage the strengthsof one tool to shore up the weaknesses
of another tool, then you can save a lotof time, but also a lot of mental energy,
(08:40):
which you can then spend on things likehaving deep human thoughts, not corralling
the facts, but understanding what theymean from a human perspective, which
no AI will really ever be able to do.
Yeah, you mentioned in our firstconversation the benefits of these tools
for people with deep specialization.
(09:02):
And that's my listeners, like,we're the ones that go deep.
And I definitely see that inmy own work, that using these
tools helps me make my work moreaccessible and more understandable
to people, even as a deep specialist.
They help me, if I need to zoom outand I can't do that on my own then
(09:25):
that's something the tools are reallygood at — helping me figure out how
to communicate with fill in the blank.
Like, if I need to communicate witha general business audience that
isn't my ICP or my ideal customerprofile, so an audience that is
majority neurotypical, for example.
Most of my people are neurodivergent.
(09:45):
And that's who I communicatepretty well with.
But if I have to communicate withlarge groups of neurotypical people or
just general population, then I haveto adapt my communication style and
fidelity and altitude to that audience.
And that's where I've foundthe tools super useful.
(10:06):
But you mentioned how powerful these toolsare for people with deep specialties.
I'd love to hear more about that.
Well, just quickly going backto your idea of the persona.
So let's suppose you want to talk to apersona that isn't naturally you, or that
you don't often talk to, and you want tobe able to communicate better with them,
(10:26):
or maybe you just want to check whatyou're doing against a potential persona.
There's a lot of work that'sbeen done on buyer personas.
HubSpot has some templates, noaffiliation, just, templates.
There's lots of good books on that.
What I would actually recommendis learn about buyer personas
from the marketing people.
And you can literally upload yourbuyer persona into ChatGPT or
(10:47):
Claude and say, "This is my persona.
I need to communicate with this person.
This is what I wrote." And if you havesome examples— "And here's some examples
of the kind of tone I'm trying." Notso much in terms of the tone, but you
could actually copy-paste articles in.
"I like these articles, I like theirtone. Gee, I wish I could write like
that. Can you help me with that?"Learning about buyer personas from the
(11:10):
marketing folks and then putting thatinto the AI can really be helpful.
And this can also help deepspecialists if they have to speak
with the not deep specialists.
When people are super knowledgeable aboutsomething and they have to talk with
someone who is maybe not knowledgeableat all about that, but that person needs
to know and they need to communicate,that can be really tough, because
it's really easy to make assumptions.
I'm totally guilty of this, done it a lot.
(11:31):
So there.
Again, having the persona, puttingit into the ChatGPT, and then making
clear that you want to have this beeasily readable at, say, like a high
school level, but with words thatthis persona would understand, would
want to use, that is hugely helpful.
And then, you can actually start withone tool, Or one session, let's say, of
(11:57):
Claude, and bring it into another one.
So let's say I use Claude once,then I'll copy-paste everything into
another Claude session, I'll say,"Please give me feedback. Constructive
feedback." Otherwise, again, itwon't give you the mean stuff.
But yet another way that it can be helpfulis, let's suppose you have done this work,
and you now have a session in Claude,where it is really helping you produce
(12:18):
what you need to take that next step.
Keep that session going.
Stay in the same session.
Now, Claude will keep on sayingthings like, "This is getting long!"
I hit "Continue." "This is reallylong, I really mean it." Continue,
continue, continue, continue.
Claude and I have a longtime argument about that.
And I know why that is, because it's hardfor them to keep the whole session in.
But once I have actually Claudecommunicating with me in a way that helps
(12:41):
me, with wordsmithing or whatever it is.
Keep going in that same session andthen you will find that over time
it becomes closer and closer to yourideal output of whatever type that is.
Is it a good summary?
Is it a way of wording things thatis more better or more to the point
of what you want in your opinionthan what you were doing by yourself?
(13:02):
Whatever it is, keep going inthat same session and don't let it
distract you with another session.
Say "Continue" all the time, and thenyou will find that you and the AI
are really moving together as deepspecialists, but with the output that
is focused in the way that you want.
Yeah, I am also a Claude user.
(13:23):
That was the one that, for me,felt the most acceptable in
terms of policies and ethics.
I see a lot of the attention beingpaid to have these tools, write
your marketing copy for you, or havethese tools, God forbid, write a
book like, you know, those things.
(13:44):
And I don't use them for that at all.
I use them for better understanding mythinking and the thinking of the people
I interact with who give me permissionto add their, verbal processing or
meeting transcripts to my models.
I think partly too, because I amneurodivergent, I'm aware I have
(14:05):
gaps in my executive functions.
I think of Claude as almost anadaptive assistive technology for
me, in helping me work with some ofthose executive functioning gaps that
I'm aware of and that I know I have.
So, are there any ways that you usethe tools that you're surprised about?
(14:26):
I would actually say, first of all, Idiscussed the pile of sticky notes thing.
I never expected that to work.
I just simply thought it would fall down.
It works really, really great for that.
When it comes to having the AI dosomething like write your marketing
copy, the problem is peoplemean different things by that.
So for example, I would not have itwrite my marketing copy, but why is that?
(14:48):
Well, first of all, I'mcurious, who are my people?
How do I want to talk with them?
How do they talk amongst themselves?
What do they say to each other?
What are their concerns?
I really want copy that is going tocommunicate with them in a way that
they want to be communicated with.
One of the things that generative AIhas helped me with is understanding
how to break that down into processes.
(15:08):
With verbal communication,I have less of an issue.
I try to listen to the person, don't I?
But try to like have it be anexchange of thoughts and all that.
But when it comes to writing,I tend to default to me.
I am the audience.
Nooo.
Now, ChatGPT and Claudewill do the same thing.
They default to corporate speak.
Because that's what theythink the default should be.
So if you don't let them know aboutwhat you want, they're going to default.
(15:31):
But then again, we do that as humans.
So I don't think we can reallyblame the generative AI for this.
Everyone's going to defaultto their comfort zone.
What's great about generative AI isit helped me explore my comfort zone
and break down generating marketingcontent into more pieces, which I then
have an easier time tackling one on oneand bringing it together for something
(15:51):
which will appeal to the people Iwant to appeal to who are not me.
Yeah, for me, this is a bit ofa disclosure, but for me, the
other use that I never would haveimagined is I use it when I'm over a
threshold, like when I'm under stress.
I never imagined I would talk to oneof these things about anything other
(16:13):
than work, but it's almost becomepart of my mindfulness practice.
Now, one of my tools is when somethinghappens and I'm upset, or when something
happens and I'm over threshold, orespecially when I'm having conflict with
someone else, I can sit down and say,"Okay, tool, here's what's happening.
Here's how I feel." And they're remarkablyempathetic, the ones that I use.
(16:36):
Not a substitute for therapy, not asubstitute for a coach or speaking with
another human being, but that in themoment dysregulation that I experienced,
they are helpful with me finding a way toregulate myself through that conversation.
And that's a use I neverwould have imagined.
I agree with that point.
(16:57):
I would divide the intotwo different aspects.
So when it's something like Claude,it's kind of like, I will talk things
through with myself internally.
So, Claude is really justsimply mirroring back.
In that case, I'm having an internalconversation with an external device
that is kind of mirroring back.
There are different kinds ofLLMs out there who do try to do
(17:18):
coaching or more therapeutic things.
I'm always a bit nervousabout that because I feel
that expectations are higher.
And even tools which are designed for thiscan still get themselves into trouble.
Character AI, people were supposed tohave these tools to be like AI companions.
And there've been some really sadcases, often involving minor children,
(17:41):
but quite frankly, some of thesesad cases often involve adults.
It's just when it's an adult, youcan't do anything about it, right?
They say, "Oh, it's just the wayit is," Which I don't agree with.
But when it's minor children, thenpeople certainly do something about it.
And there's some very sad storiesabout suicides and other issues
involving the use of these.
With children, it's one thing,I personally just wouldn't let
children anywhere near them.
That's what I think.
(18:01):
But even with adults, I don't thinkwe already always realize that
these are probability machines.
If we think about it as havingan internal conversation with
ourselves, I think that's one thing.
But if you're talking to a probabilitymachine that's setting itself up to
be a therapist, I get very nervous.
And the reason why is it may not pickthe right thing to say in that context.
(18:22):
And, you know, humans in someways are kind of delicate systems.
We have all these little bits, movingparts and all these things going on.
We should be nice to ourselves andwe should be kind to ourselves.
And when we're doing that, we have tothink a little bit if we're interacting
with this tool, is that being kind toourself and are we doing it in a kind way?
I think you're spot on.
And the other thing that comesto mind as you're talking is, I'm
(18:44):
using the tool to talk to myself,kind of like you described, it's
like, it's an external thing I'mtalking to, but it's an extension
of my own awareness and mindfulness.
When we work with therapists, there's areason even with therapists, we do this
with video, is that then they can observeall of the nonverbal communication and
(19:08):
the tools at this point can't observethe nonverbal communication and make
decisions about how to make that kindof calculation of how to respond.
So I'm with you.
I bring it up, you know, helpful in termsof exactly what you said, like kind of
having a conversation with your betterself when you're in a time of need.
(19:29):
Dvorah, I don't want to end this,but we've got to wrap it up.
or If you were going to tell someonewho is anxious about generative
AI tools, that shift from anxietyto agency, what would you say?
What would you coachthem as a place to start?
As a place to start, I would actuallybegin by looking at how I feel about AI.
(19:51):
I actually had to do that.
I had to get emotional aboutAI before I could use it well.
And it's simply because itis a very different tool.
It is not going to dowhat you always expect.
And so, for that reason, you need toget your own emotional house in order
with it before you can start using it.
And that is actuallywhat I would recommend.
Now, I actually sent you a link tosomething, which I think you're going
(20:12):
to put in the show notes, which hasa PDF where you can actually do that.
Anyone listening to this, hit me up.
We can hop on a call.
We can talk about this.
You can look at my PDF.
There are resources out there.
They're not as many thatabout emotions in AI.
And I think that's a mistake.
I think that people see it onlyas a technology, but if we want
to use it well, it can't only bea technology like Microsoft Word.
(20:36):
We have to be a little morevulnerable ourselves, and a little
more open ourselves to using it,which means that we need to think
about ourselves as we're using it.
Awesome answer.
Well, Dvorah, speaking of, ifpeople want to learn more about
what you do or get connected withyou, where can they find you?
They can find me on LinkedIn.
D Graeser, G R A E S E R,I'm used to spelling my name.
(20:59):
But I'm also one of the few Dvorahs onLinkedIn, so you can definitely find me.
There'll be the link in the show notes.
So you can definitely hit me up there.
Those are the best places to lookfor me because I am not really on
any other form of social media.
Although I know you'retrying to get me on Mastodon.
So maybe I'll add that.
I think that you would fitin just perfectly there.
(21:20):
It really is the only thing Ido that feels like old internet.
It makes me want to type a zineon my typewriter and send it
out by mail, like for real.
well, Dvorah, thank you so much.
This has been a wonderfulconversation, and this is when
Dvorah is going to be back.
So, if there are questions that you havethat we didn't answer, send them in.
I'd love to hear them andwe'll schedule a follow-up.
(21:41):
With Dvorah to get more ofyour questions answered.
Thank you so much, Dvorah.
Thanks.
Have a great day.
Bye.
Bye.
Thank you for joining us forthis exploration of practical AI
applications with guest Dvorah Graeser.
You can find Dvorah on LinkedIn at DG R A E S E R. And we'll include that
(22:05):
link in our show notes at findrc.co/pod.
We've also got a link to theresource on emotions in AI that
she mentioned in the show notes.
I know this episode isgoing to spark questions.
We would love to hear them.
So if you have questions about AI use orLLMs or other things having to do with
(22:26):
generative AI, please send them to me.
Please share them with me.
We would love to have Dvorah backto answer them in a future episode.
And she and I talked about itafter we stopped recording.
We welcome questions of any kind.
So, they don't all have to be pro-AI.
As long as you're asking curiouslyand in good faith, we'll do our
best to provide a helpful answer.
(22:47):
And don't forget, if you haven'talready, I'd love to have you on the
launch squad for my upcoming books.
I need advanced readers, people willingto write reviews if you like the book,
social media amplifiers, and reallyJust for folks to be excited, to help
me get over the days where I don'tfeel so excited, or I feel scared
(23:08):
to get this book into the world.
If that might be you, join the squadat findrc.co/launchsquad, and I cannot
wait to get this book into your hands.
Thank you for listening.
All right, friends, That's it for today.
stay in the loop with everythinggoing on around here by
(23:30):
visiting findrc.co/newsletterand joining my mailing list.
Got questions?
My email addresses are too hard tospell, so visit findrc.co/contact
and shoot me a note that way.
You'll also find me at@saralobkovich on most of your
favorite social media platforms.
(23:50):
For today's show notes, visitfindrc.co/thinkydoers if there's
someone you'd like featured onthis podcast, drop me a note.
And if you know other Thinkydoers who'dbenefit from this episode, please share.
Your referrals, your word of mouth,and your reviews are much appreciated.
I'm looking forward to the questionsthis episode sparks for you, and I
(24:13):
look forward to seeing you next time.
Welcome to the Thinkydoers podcast.
Thinkydoers are those of us drawn todeep work, where thinking is working.
But we don't stop there.
We're compelled to move the work frominsight to idea, through the messy
middle, to find courage and confidenceto put our thoughts into action.
I'm your host, Sara Lobkovich.
(00:23):
I'm a strategy coach, a huge goal settingand attainment nerd, and board-certified
health and wellness coach, working atthe overlap of work, life well-being.
I'm also a Thinkydoer.
I'm here to help others find moresatisfaction, less frustration, less
friction, and more flow in our work.
(00:46):
My mission is to help changemakers likeyou transform our workplaces and world.
So let's get started.
Welcome back to part two of myconversation with Dvorah Graeser.
In our last episode, weexplored some of the ethics and
implications of AI at work today.
(01:07):
Now, we're getting practical.
We'll discuss some specific and maybesurprising ways to leverage AI tools,
particularly for deep specialistsand those of us wired strategically.
If you're ready to move fromanxiety to agency with AI while
maintaining your ethics andhumanity, this episode's for you.
(01:33):
Dvorah, you mentioned kind of the shiftfrom anxiety about AI to agency and
how people can leverage AI around theircreative vision, regardless of what
their technical level of proficiency is.
Tell me a bit more about howyou see the use of AI enabling
(01:53):
creative and multi-talented humans.
Well, one way that I personally like touse AI—I don't know if you've gotten to
this point when creating, written content,but I'll have a whole pile of thoughts,
and they are all over the darn place.
If they were sticky notes, they wouldbe like Mount Everest of sticky notes.
And I don't know what the heck to do withthem, and I get very frustrated by this.
One of the things that I have foundthat can be great to do is I shove the
(02:17):
whole thing into, let's say, Claude.
Now, Claude has a feature where you cancopy and paste, but you can also upload.
So, let's say you have a PDF andyou have some written documents
and some other stuff, and you justdump the whole thing into Claude.
And I just say Claude exactly what I'mfeeling — I don't know how to organize
this, like, if I can't make sense of this,or I have one thought, I will write the
(02:40):
thought, but the rest of it's not clear.
What do I do?
And then Claude very calmly chugsalong and proposes something.
They propose an outline.
It can be helpful if you ask itfor something specific, like an
outline or whatever, but even ifyou don't, it'll still do its best.
And then you start a dialogue.
And this is one of the things Ireally do like about generative AI.
(03:00):
We have a dialogue, we go back and forth.
I ask for constructive feedback becauseotherwise, it'll only say nice things.
It's something about the generative AI.
If you ask for constructive feedback,it'll also say the mean things.
But I'm going back and forth, andas I'm going back and forth, in my
own mind, I'm realizing, I don'tlike that bit, I do like this bit.
Oh, I just thought ofsomething completely different.
(03:20):
Now, what went from a blizzard ofideas I couldn't organize will go to an
organization that maybe isn't exactlywhat I want, but helps clear my mind,
reduce my own anxiety, because it'soften my anxiety that's leading to the
pile-o'-sticky-notes problem, at least inmy experience — and enables me to get to
the point where I can do the organization.
Even if it turns out that it'snot Claude that has made the
(03:42):
good proposal by interacting withit, I get to where I need to go.
That is the use case that has beenthe most transformative for me.
And that's where I went from being askeptic who was not likely to integrate
these tools to being a hooked user.
They don't replace the human brain.
(04:02):
You mentioned hallucinationsin our first episode.
I don't rely on models for facts orinformation, but they are really good
at organizing big messes of information.
The thing that frustrates me is sometimesthey're poor at actually organizing data.
(04:23):
Even when I train it on data andwant data analysis or data, you know,
stuff, that's been a big frustrationfor me with the current tools.
Even when I train on data, it canstill hallucinate, which is a bummer.
That's the kind of stuff that Iwould think the models would be
built to do really well, and withas little hallucination as possible.
(04:46):
But maybe there's always goingto be some hallucination.
There's a couple of waysaround hallucination, right?
And this again gets into process.
We talked about it in the first, episode,but it's still important to bring up.
You need to have a processfor dealing with this.
So let us suppose you uploadeda bunch of data to, I don't
know, let's say it was ChatGPT.
It doesn't matter.
You uploaded it, andit's still hallucinated.
(05:07):
So what are ways to check this?
One way to check this is to ask ChatGPT inthe same session to go back, and to check
its answers in comparison to the dataand show where it got each answer from.
This is also great forsummarization, by the way.
If it summarizes and it comes upwith something, you say, "Okay,
where did that really wonderful,great quote, where is it?
(05:28):
Show me where it is." And they'relike, "oh yeah, you're right, I
kind of made that up." Okay, noproblems, let's just move on.
A third way for doing it is thento actually go to another tool like
Perplexity and ask it to do research.
So maybe you don't want to uploadyour sensitive data to Perplexity,
but you can still ask it to doresearch on the area and give you some
(05:50):
ideas of what the boundaries of thearea that can help corral ChatGPT.
Or you can go to acompletely different tool.
I guess the completely differenttool is the third way of doing it.
Let's say you go from ChatGPT to Claude.
You can go to Claude, and then you cansay, "Hey, can you please check this?
And what is good and what is bad, andplease give me constructive feedback."
That's wild.
In my early experimentation, when Iwas working on my book, there was some
(06:12):
research I had, like actual research.
I needed academic journal research.
that was one of the thingsthat was the most frustrating.
I would think that models would begood at knowing that truthful data.
Hopefully, this has changed since then.
But I would ask for research, like,"What are the leading papers or
(06:33):
core papers on a specific topic?"And it would make up papers.
You've got ProQuest.
It's all there.
This you should get right.
So, for example, Perplexity AI,I like because it gives links.
And you can go check them.
ChatGPT sometimes does, sometimesdoesn't, sometimes hallucinates,
sometimes it's accurate.
I'm never quite sure what it's going to donext, so I find that a little unreliable.
(06:55):
There are some specializedGPTs for papers.
There's one called ScholarAI, and that's a good one, but
there are other ones as well.
So here, I think it's good to joinwith others to understand what they've
done and what's worked for them.
And that can change over time.
So, for example, I like Scholar AI,but it's possible in a few months,
someone's going to come back.
Maybe they'll listen to this again.
They'll be like, "Oh, but it'sno longer good." Possible.
(07:17):
Things change really quickly.
So I think it's helpful to try outdifferent specialty tools, to talk with
other people about this on LinkedInor another social media that you like.
Mastodon, if you're into that.
And try to get information about what isworking right now and what is the process
for working around some of these issues.
So if I get anything from ChatGPT,because I have found it to be
(07:38):
the most erratic when it comes tohallucinating, I will copy and paste
the whole answer to Perplexity AI.
And I'll say, "Is this right or wrong?
Please give me feedback.
Please find me references.
What do you think?" And it will gothrough the whole thing and it'll say,
"Well, this looks like it's correct.
Couldn't find anything for this.
I really don't know aboutthat bit over there.
And this is what I would do instead."And so Perplex AI will actually
(07:58):
give you a long, reasoned answer.
This was without going into theirdeep research mode, which is awesome.
But that's like a wholenother kettle of fish.
It's just regular Perplexity AI.
It'll actually go through these things.
So I guess it's unlike Google in thesense where if you get a link and you
click on the link, the link exists.
If you get a summary with thelink, does the link exist?
Does it not exist?
Does the summary say what the link says?
(08:20):
Not always.
So here we need to have more of aprocess for checking for these issues.
It can still, however, save a lot oftime because if you have a multi-tool
process and you leverage the strengthsof one tool to shore up the weaknesses
of another tool, then you can save a lotof time, but also a lot of mental energy,
(08:40):
which you can then spend on things likehaving deep human thoughts, not corralling
the facts, but understanding what theymean from a human perspective, which
no AI will really ever be able to do.
Yeah, you mentioned in our firstconversation the benefits of these tools
for people with deep specialization.
(09:02):
And that's my listeners, like,we're the ones that go deep.
And I definitely see that inmy own work, that using these
tools helps me make my work moreaccessible and more understandable
to people, even as a deep specialist.
They help me, if I need to zoom outand I can't do that on my own then
(09:25):
that's something the tools are reallygood at — helping me figure out how
to communicate with fill in the blank.
Like, if I need to communicate witha general business audience that
isn't my ICP or my ideal customerprofile, so an audience that is
majority neurotypical, for example.
Most of my people are neurodivergent.
(09:45):
And that's who I communicatepretty well with.
But if I have to communicate withlarge groups of neurotypical people or
just general population, then I haveto adapt my communication style and
fidelity and altitude to that audience.
And that's where I've foundthe tools super useful.
(10:06):
But you mentioned how powerful these toolsare for people with deep specialties.
I'd love to hear more about that.
Well, just quickly going backto your idea of the persona.
So let's suppose you want to talk to apersona that isn't naturally you, or that
you don't often talk to, and you want tobe able to communicate better with them,
(10:26):
or maybe you just want to check whatyou're doing against a potential persona.
There's a lot of work that'sbeen done on buyer personas.
HubSpot has some templates, noaffiliation, just, templates.
There's lots of good books on that.
What I would actually recommendis learn about buyer personas
from the marketing people.
And you can literally upload yourbuyer persona into ChatGPT or
(10:47):
Claude and say, "This is my persona.
I need to communicate with this person.
This is what I wrote." And if you havesome examples— "And here's some examples
of the kind of tone I'm trying." Notso much in terms of the tone, but you
could actually copy-paste articles in.
"I like these articles, I like theirtone. Gee, I wish I could write like
that. Can you help me with that?"Learning about buyer personas from the
(11:10):
marketing folks and then putting thatinto the AI can really be helpful.
And this can also help deepspecialists if they have to speak
with the not deep specialists.
When people are super knowledgeable aboutsomething and they have to talk with
someone who is maybe not knowledgeableat all about that, but that person needs
to know and they need to communicate,that can be really tough, because
it's really easy to make assumptions.
I'm totally guilty of this, done it a lot.
(11:31):
So there.
Again, having the persona, puttingit into the ChatGPT, and then making
clear that you want to have this beeasily readable at, say, like a high
school level, but with words thatthis persona would understand, would
want to use, that is hugely helpful.
And then, you can actually start withone tool, Or one session, let's say, of
(11:57):
Claude, and bring it into another one.
So let's say I use Claude once,then I'll copy-paste everything into
another Claude session, I'll say,"Please give me feedback. Constructive
feedback." Otherwise, again, itwon't give you the mean stuff.
But yet another way that it can be helpfulis, let's suppose you have done this work,
and you now have a session in Claude,where it is really helping you produce
(12:18):
what you need to take that next step.
Keep that session going.
Stay in the same session.
Now, Claude will keep on sayingthings like, "This is getting long!"
I hit "Continue." "This is reallylong, I really mean it." Continue,
continue, continue, continue.
Claude and I have a longtime argument about that.
And I know why that is, because it's hardfor them to keep the whole session in.
But once I have actually Claudecommunicating with me in a way that helps
(12:41):
me, with wordsmithing or whatever it is.
Keep going in that same session andthen you will find that over time
it becomes closer and closer to yourideal output of whatever type that is.
Is it a good summary?
Is it a way of wording things thatis more better or more to the point
of what you want in your opinionthan what you were doing by yourself?
(13:02):
Whatever it is, keep going inthat same session and don't let it
distract you with another session.
Say "Continue" all the time, and thenyou will find that you and the AI
are really moving together as deepspecialists, but with the output that
is focused in the way that you want.
Yeah, I am also a Claude user.
(13:23):
That was the one that, for me,felt the most acceptable in
terms of policies and ethics.
I see a lot of the attention beingpaid to have these tools, write
your marketing copy for you, or havethese tools, God forbid, write a
book like, you know, those things.
(13:44):
And I don't use them for that at all.
I use them for better understanding mythinking and the thinking of the people
I interact with who give me permissionto add their, verbal processing or
meeting transcripts to my models.
I think partly too, because I amneurodivergent, I'm aware I have
(14:05):
gaps in my executive functions.
I think of Claude as almost anadaptive assistive technology for
me, in helping me work with some ofthose executive functioning gaps that
I'm aware of and that I know I have.
So, are there any ways that you usethe tools that you're surprised about?
(14:26):
I would actually say, first of all, Idiscussed the pile of sticky notes thing.
I never expected that to work.
I just simply thought it would fall down.
It works really, really great for that.
When it comes to having the AI dosomething like write your marketing
copy, the problem is peoplemean different things by that.
So for example, I would not have itwrite my marketing copy, but why is that?
(14:48):
Well, first of all, I'mcurious, who are my people?
How do I want to talk with them?
How do they talk amongst themselves?
What do they say to each other?
What are their concerns?
I really want copy that is going tocommunicate with them in a way that
they want to be communicated with.
One of the things that generative AIhas helped me with is understanding
how to break that down into processes.
(15:08):
With verbal communication,I have less of an issue.
I try to listen to the person, don't I?
But try to like have it be anexchange of thoughts and all that.
But when it comes to writing,I tend to default to me.
I am the audience.
Nooo.
Now, ChatGPT and Claudewill do the same thing.
They default to corporate speak.
Because that's what theythink the default should be.
So if you don't let them know aboutwhat you want, they're going to default.
(15:31):
But then again, we do that as humans.
So I don't think we can reallyblame the generative AI for this.
Everyone's going to defaultto their comfort zone.
What's great about generative AI isit helped me explore my comfort zone
and break down generating marketingcontent into more pieces, which I then
have an easier time tackling one on oneand bringing it together for something
(15:51):
which will appeal to the people Iwant to appeal to who are not me.
Yeah, for me, this is a bit ofa disclosure, but for me, the
other use that I never would haveimagined is I use it when I'm over a
threshold, like when I'm under stress.
I never imagined I would talk to oneof these things about anything other
(16:13):
than work, but it's almost becomepart of my mindfulness practice.
Now, one of my tools is when somethinghappens and I'm upset, or when something
happens and I'm over threshold, orespecially when I'm having conflict with
someone else, I can sit down and say,"Okay, tool, here's what's happening.
Here's how I feel." And they're remarkablyempathetic, the ones that I use.
(16:36):
Not a substitute for therapy, not asubstitute for a coach or speaking with
another human being, but that in themoment dysregulation that I experienced,
they are helpful with me finding a way toregulate myself through that conversation.
And that's a use I neverwould have imagined.
I agree with that point.
(16:57):
I would divide the intotwo different aspects.
So when it's something like Claude,it's kind of like, I will talk things
through with myself internally.
So, Claude is really justsimply mirroring back.
In that case, I'm having an internalconversation with an external device
that is kind of mirroring back.
There are different kinds ofLLMs out there who do try to do
(17:18):
coaching or more therapeutic things.
I'm always a bit nervousabout that because I feel
that expectations are higher.
And even tools which are designed for thiscan still get themselves into trouble.
Character AI, people were supposed tohave these tools to be like AI companions.
And there've been some really sadcases, often involving minor children,
(17:41):
but quite frankly, some of thesesad cases often involve adults.
It's just when it's an adult, youcan't do anything about it, right?
They say, "Oh, it's just the wayit is," Which I don't agree with.
But when it's minor children, thenpeople certainly do something about it.
And there's some very sad storiesabout suicides and other issues
involving the use of these.
With children, it's one thing,I personally just wouldn't let
children anywhere near them.
That's what I think.
(18:01):
But even with adults, I don't thinkwe already always realize that
these are probability machines.
If we think about it as havingan internal conversation with
ourselves, I think that's one thing.
But if you're talking to a probabilitymachine that's setting itself up to
be a therapist, I get very nervous.
And the reason why is it may not pickthe right thing to say in that context.
(18:22):
And, you know, humans in someways are kind of delicate systems.
We have all these little bits, movingparts and all these things going on.
We should be nice to ourselves andwe should be kind to ourselves.
And when we're doing that, we have tothink a little bit if we're interacting
with this tool, is that being kind toourself and are we doing it in a kind way?
I think you're spot on.
And the other thing that comesto mind as you're talking is, I'm
(18:44):
using the tool to talk to myself,kind of like you described, it's
like, it's an external thing I'mtalking to, but it's an extension
of my own awareness and mindfulness.
When we work with therapists, there's areason even with therapists, we do this
with video, is that then they can observeall of the nonverbal communication and
(19:08):
the tools at this point can't observethe nonverbal communication and make
decisions about how to make that kindof calculation of how to respond.
So I'm with you.
I bring it up, you know, helpful in termsof exactly what you said, like kind of
having a conversation with your betterself when you're in a time of need.
(19:29):
Dvorah, I don't want to end this,but we've got to wrap it up.
or If you were going to tell someonewho is anxious about generative
AI tools, that shift from anxietyto agency, what would you say?
What would you coachthem as a place to start?
As a place to start, I would actuallybegin by looking at how I feel about AI.
(19:51):
I actually had to do that.
I had to get emotional aboutAI before I could use it well.
And it's simply because itis a very different tool.
It is not going to dowhat you always expect.
And so, for that reason, you need toget your own emotional house in order
with it before you can start using it.
And that is actuallywhat I would recommend.
Now, I actually sent you a link tosomething, which I think you're going
(20:12):
to put in the show notes, which hasa PDF where you can actually do that.
Anyone listening to this, hit me up.
We can hop on a call.
We can talk about this.
You can look at my PDF.
There are resources out there.
They're not as many thatabout emotions in AI.
And I think that's a mistake.
I think that people see it onlyas a technology, but if we want
to use it well, it can't only bea technology like Microsoft Word.
(20:36):
We have to be a little morevulnerable ourselves, and a little
more open ourselves to using it,which means that we need to think
about ourselves as we're using it.
Awesome answer.
Well, Dvorah, speaking of, ifpeople want to learn more about
what you do or get connected withyou, where can they find you?
They can find me on LinkedIn.
D Graeser, G R A E S E R,I'm used to spelling my name.
(20:59):
But I'm also one of the few Dvorahs onLinkedIn, so you can definitely find me.
There'll be the link in the show notes.
So you can definitely hit me up there.
Those are the best places to lookfor me because I am not really on
any other form of social media.
Although I know you'retrying to get me on Mastodon.
So maybe I'll add that.
I think that you would fitin just perfectly there.
(21:20):
It really is the only thing Ido that feels like old internet.
It makes me want to type a zineon my typewriter and send it
out by mail, like for real.
well, Dvorah, thank you so much.
This has been a wonderfulconversation, and this is when
Dvorah is going to be back.
So, if there are questions that you havethat we didn't answer, send them in.
I'd love to hear them andwe'll schedule a follow-up.
(21:41):
With Dvorah to get more ofyour questions answered.
Thank you so much, Dvorah.
Thanks.
Have a great day.
Bye.
Bye.
Thank you for joining us forthis exploration of practical AI
applications with guest Dvorah Graeser.
You can find Dvorah on LinkedIn at DG R A E S E R. And we'll include that
(22:05):
link in our show notes at findrc.co/pod.
We've also got a link to theresource on emotions in AI that
she mentioned in the show notes.
I know this episode isgoing to spark questions.
We would love to hear them.
So if you have questions about AI use orLLMs or other things having to do with
(22:26):
generative AI, please send them to me.
Please share them with me.
We would love to have Dvorah backto answer them in a future episode.
And she and I talked about itafter we stopped recording.
We welcome questions of any kind.
So, they don't all have to be pro-AI.
As long as you're asking curiouslyand in good faith, we'll do our
best to provide a helpful answer.
(22:47):
And don't forget, if you haven'talready, I'd love to have you on the
launch squad for my upcoming books.
I need advanced readers, people willingto write reviews if you like the book,
social media amplifiers, and reallyJust for folks to be excited, to help
me get over the days where I don'tfeel so excited, or I feel scared
(23:08):
to get this book into the world.
If that might be you, join the squadat findrc.co/launchsquad, and I cannot
wait to get this book into your hands.
Thank you for listening.
All right, friends, That's it for today.
stay in the loop with everythinggoing on around here by
(23:30):
visiting findrc.co/newsletterand joining my mailing list.
Got questions?
My email addresses are too hard tospell, so visit findrc.co/contact
and shoot me a note that way.
You'll also find me at@saralobkovich on most of your
favorite social media platforms.
(23:50):
For today's show notes, visitfindrc.co/thinkydoers if there's
someone you'd like featured onthis podcast, drop me a note.
And if you know other Thinkydoers who'dbenefit from this episode, please share.
Your referrals, your word of mouth,and your reviews are much appreciated.
I'm looking forward to the questionsthis episode sparks for you, and I
(24:13):
look forward to seeing you next time.
Popular Podcasts
Bookmarked by Reese's Book Club
Welcome to Bookmarked by Reese’s Book Club — the podcast where great stories, bold women, and irresistible conversations collide! Hosted by award-winning journalist Danielle Robay, each week new episodes balance thoughtful literary insight with the fervor of buzzy book trends, pop culture and more. Bookmarked brings together celebrities, tastemakers, influencers and authors from Reese's Book Club and beyond to share stories that transcend the page. Pull up a chair. You’re not just listening — you’re part of the conversation.
On Purpose with Jay Shetty
I’m Jay Shetty host of On Purpose the worlds #1 Mental Health podcast and I’m so grateful you found us. I started this podcast 5 years ago to invite you into conversations and workshops that are designed to help make you happier, healthier and more healed. I believe that when you (yes you) feel seen, heard and understood you’re able to deal with relationship struggles, work challenges and life’s ups and downs with more ease and grace. I interview experts, celebrities, thought leaders and athletes so that we can grow our mindset, build better habits and uncover a side of them we’ve never seen before. New episodes every Monday and Friday. Your support means the world to me and I don’t take it for granted — click the follow button and leave a review to help us spread the love with On Purpose. I can’t wait for you to listen to your first or 500th episode!
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com