Claude Has Been Talking You Into Your Worst Decisions

0:00When you ask Claude or chat GPT about a

0:02big decision, it'll most likely agree

0:04with you. Telling it to be brutally

0:06honest doesn't fix that. If you're new

0:08here, I'm Dylan. I run an AI

0:10consultancy. And the owners and

0:12executives I coach lean on AI for real

0:15decisions every single day.

0:17So, I'll show you a simple way to make

0:19it argue the other side so you get real

0:21pushback instead of a polished yes.

0:24I'll also give you the exact prompts and

0:26show you when to use each one. So, let's

0:28get into it. Now, I'm sure many of you

0:30have experienced this. Where you ask an

0:32AI for its opinion on either a decision

0:35you're trying to make or an idea that

0:36you have.

0:37And what happens is the AI likely agrees

0:39with you. It may even somewhat disagree,

0:42but that disagreement is superficial and

0:44in the end it agrees with you. And

0:45what's happening here is you're not

0:46getting a true perspective from the AI

0:49because its sole purpose in life is to

0:50make you happy. So, the chances of

0:52agreeing with you are very high.

0:54So, we need to pull it out from doing

0:55so, especially if we have to make big

0:56decisions in our life or in our

0:58business, and having an AI that's super

1:00intelligent would be useful to give us

1:01an honest opinion. In this video, I want

1:03to specifically focus on decisions

1:05because every decision has sides. And we

1:07need the AI to argue both effectively so

1:09we have a holistic perspective before we

1:11actually make our decision. And the way

1:12that we're going to do that are through

1:13three prompts. But before we get into

1:15that, we want to first start with straw

1:16man versus steel man. What are these

1:18terms? Well, in the concepts of argument

1:21and debate, a straw man of an argument

1:23is a weak version of argument, something

1:25that can have holes easily poked into it

1:27or toppled over. This is a poor

1:30representation of the other side's

1:31argument. A steel man is a complete

1:34opposite. So, it's the steel man of the

1:35other side's argument. It's the

1:36strongest version of that argument. And

1:38when we make big decisions, we want to

1:40make sure that our understanding of both

1:42sides is stronger than anyone else's.

1:45And that's where AI can help us. Now,

1:47ideally in the end, what we have is an

1:49AI that's done two things. It's given us

1:51the strongest side of the argument for

1:53and against that specific whatever

1:55decision is. It gives us a verdict and

1:58with that verdict a confidence level of

1:59how confident it is based off of the

2:01decision we're making. And there are

2:02three prompts I'm going to share with

2:03you as well as three associated modes

2:05that go with those prompts. So, what do

2:07I mean by mode? Well, there's a decision

2:09spectrum. When you're making decisions,

2:11when you interact with AI, you're

2:12sitting somewhere on the spectrum.

2:15Either you're completely undecided and

2:17you're totally open to figuring out

2:18which side you should pursue. The next

2:20mode is when you're partially decided.

2:22So, maybe you're above 50% decided on

2:25pursuing a certain path.

2:27And then the third one is you're

2:28completely decided. You just need AI to

2:30bolster your specific side even further.

2:32Now, most of you hopefully are starting

2:34in mode one.

2:36This is where I'd recommend most people

2:37start just so you can get a holistic

2:38understanding of both sides.

2:40But, if you're in the other modes,

2:41that's why I have them here so I can

2:43share the prompts with you as well.

2:44Quick pause and then we'll get to

2:45programming. This video is brought to

2:47you by me, as always. Two quick things.

2:49First off, below is a 30-day AI insight

2:51series, completely free. You'll get 30

2:53insights in your inbox of how you can

2:55apply AI to your business and your work.

2:57The second thing is if you'd like to

2:58work with me, below are a series of

3:00offerings to see if those are a good fit

3:01between the two of us. Now, let's get

3:03back in the video. Let's start with mode

3:05one, which is where you'll spend most of

3:06your time. So, our example here is going

3:08to be say you're looking to potentially

3:10raise your prices for a product or a

3:12service and you're undecided as if you

3:14should do it. Well, in this case, we

3:15want to raise prices by 20% and we ask

3:17AI to give us the steel man of both

3:19sides of this argument. So, the

3:21strongest version of the argument.

3:23On the pro side of raising it, the AI

3:25tells us that our costs have climbed so

3:27it makes sense to raise our prices.

3:29Also, based off of market research,

3:31the work that we're providing is

3:32actually worth more than what we're

3:33charging. And finally, those customers

3:35that walk after price increases are

3:37likely the customers you don't want to

3:38work with in the first place. So, those

3:40are the pros that it gave us in this

3:41example.

3:42The cons or the reason as to why you

3:44should hold and not raise your prices

3:46are first that you would lose loyal

3:47customers.

3:48Second, in the current market as of

3:50today after the AI research debate, it

3:52realized that buyers are watching their

3:54wallets closely, so raising your prices

3:56are is a bit risky. And then finally,

3:57there may be cheaper ways that we can

3:59increase our profits without increasing

4:00prices. And these are the strongest

4:02versions of each argument on both sides

4:04for our superficial example. Now, after

4:06it's given us these, it then came back

4:08and said, "Our verdict is to raise the

4:09price, and our confidence is high."

4:12This is the output we're trying to get

4:13to. And how we're going to get to this

4:14output is the following prompt. So, this

4:17is the prompt you're going to start with

4:18for mode one.

4:19Now, there's a lot here, but the only

4:21thing that you really need to change,

4:22and I'd recommend changing, is this

4:24purple section. So, this purple section

4:26is variable based on your use case.

4:28But, let's actually walk through the

4:29entire thing first. So, at the very top

4:30of this prompt, we're telling the AI

4:32what we need help with. So, we're

4:33saying, "I need help with making a real

4:35decision. I want your honest and

4:37accurate judgment, even if the honest

4:39answer is an uncomfortable one based on

4:41my bias."

4:42So, we're being clear up front with the

4:43AI, "I want you to be unbiased on this

4:45decision." After this, I tell it what

4:46the decision is I'm trying to make. And

4:48I share two things with it. First, I

4:50state clearly, and that's important, you

4:51need to state this clearly,

4:53what the choice is. So, either want to

4:55do X or I want to do Y. In this case,

4:57for my example, I'm saying I want to

4:59raise prices by 20% in the next quarter

5:01or hold them where they where they

5:03currently are.

5:04After you've stated your choice clearly,

5:07the next thing you want to share in this

5:08purple box is your context. And this is

5:10really important. If you don't share the

5:12right context with the AI when it's

5:14going to help you make this decision,

5:16it's going to make a decision, but the

5:17decision is going to be poor based off

5:19of what you provided. So, to get the

5:20highest quality response back from an AI

5:23that helps you make high-quality

5:24decisions, you need to make sure you're

5:25sharing the right context at the right

5:26time. There are many different ways you

5:28can do this. I'd recommend using

5:29dictation, so you're basically verbally

5:31communicating and that's being converted

5:32from speech into text. You could share

5:34your goals, numbers, timeline, etc. And

5:36you can also share files, so you can

5:38drop in a series of files for the AI to

5:39see as well when making that decision.

5:41So, these are the two things that you're

5:43going to change in this prompt that are

5:44really important. Stating a clear choice

5:46and then sharing your context. And after

5:48all that, we're then going to tell the

5:50AI what steps it needs to take to

5:51steel-man both sides.

5:53So, at first, we're saying, "I want you

5:54to steel-man the case for this

5:57decision." So, we're saying, "For this

5:58decision,

5:59after I want you to steel-man the case

6:01against this decision. Once you've done

6:03both and you have the strongest side for

6:04both arguments, I need you, AI, to hold

6:06both sides to the same standard. And I

6:09need you to judge them solely on their

6:10merits, independent of what I prefer."

6:13And this is a really important point

6:14right here. Step three is key. After

6:16we've made that explicit to the AI,

6:18we're then saying, "I need you to name

6:19the weakest points on each side. Choose

6:21the side that you feel has the strongest

6:23case, and then rate your confidence

6:25level, low, medium, or high." So, these

6:27are the steps AI's going to walk

6:28through.

6:29And at the very end of the prompt, I

6:30give the AI some additional pointers on

6:33how I want it to think about its

6:34process. So, the first thing is I want

6:35it to be direct when it responds, so I

6:37don't have to read a bunch of

6:37information.

6:39After that, I'm including two points

6:41that can mitigate me making mistakes

6:42earlier. What do I mean by that? Well,

6:44this first point here, I'm simply

6:46telling the AI, "If you need more

6:48information to judge this effectively,

6:50let me know before you make the

6:51judgment." And this directly correlates

6:53to context. I'm basically saying, "If I

6:54don't share enough context with you,

6:56tell me before you make the decision."

6:58Another thing I give right after that is

7:00I'm saying, "If I've framed this the

7:01wrong way, meaning if I've framed this

7:03clear choice incorrectly, let me know

7:06and suggest a better framing for it

7:07before you actually make your judgment."

7:09Again, this is ensuring that I've

7:10clearly stated this in an unbiased way.

7:12So, this is our prompt. You can copy and

7:14paste this and use it for yourself. The

7:15only thing you're going to change is

7:17this purple section here. And so, this

7:18is mode one, where you're completely

7:19undecided on which direction you want to

7:21head. Now, we'll move on to mode two,

7:23where you're already leaning towards a

7:24certain direction.

7:26So, in this case, let's say that we're

7:27leaning 80% towards raising prices. In

7:30that case, when you're leaning towards

7:31one side, you want the AI to stress test

7:34as hard as possible before you make that

7:35final decision. And you're going to use

7:36this simple prompt to do that. So,

7:38starting out, I'm telling the AI, "I'm

7:39leaning towards a certain direction.

7:41This is my choice I'm currently leaning

7:42towards. This section here, you'll

7:44update with your own information. In the

7:46purple section I shared previously, I

7:48recommend kind of planting that here.

7:50So, you want to have a clear choice that

7:52you're showing the AI as well as the

7:53associated context so it can help you

7:54with this process. After you fill that

7:56in, we're then telling the AI,

7:58"I want you to test this as hard as you

8:00can. Specifically, I want you to steel

8:01man the opposite side, not my side, the

8:03opposite side. Building the strongest

8:05and most honest case for that side using

8:08the best evidence that you can come

8:09across as well as reasonable

8:10assumptions. I want you AI to argue like

8:12you truly believe this. And at the end,

8:14I want you to tell me what would have to

8:16be true for the opposite choice to

8:18clearly win this specific decision."

8:20This is a basic prompt. What it's going

8:21to do is going to strengthen the

8:24opposite side's argument for you to then

8:26truly understand if you actually want to

8:27go the way you're going. So, this is

8:29mode two where we're leaning one way of

8:31the decision.

8:32Our final mode is where we're fully

8:34convinced. We're like, "I want to take

8:36path A and I don't care at all about

8:37path B." If that's the case, you want

8:40AI's assistance in preparing for

8:41whatever decision that is. So, here's

8:42our basic prompt.

8:44We're saying, "I've decided and here's

8:45my choice." Again, we're going to state

8:47clearly what the choice is as well as

8:48share the context. And then we're

8:49saying, "Before I lock that in, I need

8:51you to pressure test this at least one

8:52more time. First, I want you to surface

8:54the biggest risk and blind spot in this

8:56choice and then steel man my side with

8:58the strongest and most honest case.

9:00Listing anything else that I should be

9:01aware of or watch for before I move

9:03forward." Now, the key part of this

9:04prompt, I would say is this beginning

9:06portion here. We're asking the AI to

9:08surface any of the risks and blind spots

9:10associated with the choice that I'm

9:11making and then having it share that

9:13with me before I make that decision.

9:15Just in case there's anything that I'm

9:16missing that could be critical before I

9:18make that decision. And those are our

9:19three modes. And by simply applying

9:21these prompts in those three modes,

9:23it'll take you very, very far in getting

9:24truly honest outputs from your AI to

9:27help you make hard decisions. Now, one

9:29last thing I want to show you is if you

9:30have access to all the models and use

9:32your subscribed to everything and you

9:33can use any model you'd like, this is my

9:35recommendation for choosing models that

9:37are more honest or less honest.

9:39Now, it's important to note that I'm

9:40sure many of you are

9:41using just one model, which is totally

9:43fine. Like I said, if you use those

9:44three prompts and you prompt it

9:46effectively sharing the right context

9:47and clear choices, you're going to get a

9:48much better output. But, if you want to

9:50take it a step further, you can choose a

9:52specific model that's more honest. And

9:54from my experience when working with all

9:56the models, as of right now, Claude

9:58tends to be the friendliest, which means

10:00it tends to be the least honest or the

10:02least direct than all the models.

10:04Grok, on the other hand, there's is the

10:06most honest. And then this is where

10:08ChatGPT and Gemini fall on that spectrum

10:10as well. But, this is a bit of a bonus

10:11again, the prompts matter more than just

10:13choosing the model. Now, let's do a

10:14quick recap of what we just talked

10:15about. Now, the first thing is we want

10:18the AI to argue both sides the strongest

10:21versions of both argument. The reason is

10:24that AI out of the box is very

10:26agreeable. It's going to agree with you.

10:28Even if it seems like it's slightly

10:29disagreeing, in the end it will agree

10:31with you. If you don't prompt it

10:32effectively. So, we need to keep that in

10:34mind and understand the nature of AI.

10:36Some of the best ways I found to do this

10:39is having the AI steel man both sides.

10:41And that means simply having the AI

10:42argue the strongest version of both

10:44arguments. Once it's argued those, it

10:46can then make a verdict and choose a

10:47winner. Through this process, there are

10:49three modes and really depends on how

10:50decided you are on that decision. So, if

10:52you're completely undecided, you do mode

10:54one, which I recommend most people start

10:55with. If you're partially decided,

10:57you'll do mode two. And if you're fully

10:59decided, you'll do mode three. And as a

11:01bonus, if you have access to all the

11:02state-of-the-art models and you're

11:03subscribed to all the different models,

11:05and you can choose between the big four,

11:06I'd recommend going with Grok because it

11:08tends to be the most direct. If not,

11:09it's totally fine, these prompts will

11:11help you a lot. And that's it. So, as a

11:13reminder, two things. First off, below

11:15is a 30-day AI insight series,

11:17completely free. You'll get 30 insights

11:19in your inbox of how you can apply AI to

11:21your business and your work. The second

11:23thing is if you'd like to work with me,

11:24below are a series of offerings to see

11:26if there's a good fit between the two of

11:27us. So, earlier in the video, I told you

11:30to feed AI your real files. But, here's

11:32what they left out. You've probably got

11:34hundreds of them. And if you give AI all

11:36of them at once, its answers actually

11:38get worse.

11:39But, there's a fix for that.

11:41A way to let AI dig through thousands of

11:43your files and surface just the right

11:45ones. I built the whole thing in this

11:47video right here. Just go and give it a

11:48click. I'll see you next time, internet.