Original subtitles

In this video,

we're going to talk about your role in using generative AI as a thought partner.

We've talked about the importance of critical thinking, but

let's be a little bit more specific.

The output from generative AI is based on the inputs on which the models

are trained, and it can recombine these inputs in ways that are not valuable,

not true, and even harmful to people.

So I always try to do five key things whenever I use generative AI, and

I'll share those five things in this short video.

So, as we've said,

working with large language models as a thought partner is very valuable.

And as these models get more powerful, it will become even more valuable.

But these models, at least today, they're not perfect and

they're often flat out wrong.

And so you have an important role to play in the way that you interact with these

models.

I mean, you can see I'm trying to depict here,

you've got to stay in control of the conversation, and

you cannot just blindly accept what comes out of these models.

So five action verbs that I always try to take when I'm using these models.

Number one is to reflect,

I do not take what comes out of these models at face value.

I always think about, I say, what do I think of that?

Does that make sense to me?

Does that match my intuition?

Is that something that seems true based on my experience?

And then not only does it seems true, but

number two is validating that things are true.

So if you're looking for factual things, not just conceptual ideas,

make sure that you validate those things.

I mean, I frankly do not use large language models for factuality, for

searches very often, unless it's with a search engine that will give

me a generated response based on underlying web pages, where I can go to

the web page and I can myself decide whether I find that web page credible.

Another thing that's happening, by the way,

is a lot of web pages are being generated by AI.

Many of these are not true, so even though it looks like your search results

are grounded in a web page that maybe is authoritative,

that web page is not necessarily so.

As usual, it's just good practice, evaluate your sources.

And with a large language model, it's hard to know what the sources are.

So be very careful about factuality,

and make sure that you validate anything that you deem to be true.

Especially if the information that you're going to be using is going to be put into

a high stakes decision,

make sure it's true before you actually base your decision on that information.

All right, third big action, debate.

Don't just be passive, if the language model tells you something, challenge it.

Let it challenge you, you challenge it.

Now, one of these you'll find is a lot of these models, they'll just fold.

If you say, I disagree with that, they'll say, yeah, you're right.

So, try to frame your questions as challenges that are kind of open ended so

that it just doesn't automatically agree with you.

The fourth action is to filter.

One of the things that these generative models are really good at is generating

lots of options.

And one of the things I like to do is I like to ask it for way more than I'm

needing, because then I can sift through it and pick the pieces that I like.

So instead of saying, give me a recommendation for how to put a title on

top of this paragraph, I'll say, give me five recommendations for how to put

a title on top of this paragraph, and then I can pick the one that I like.

But filtering is really valuable because ,a, the generative AI model can give

you a lot of options, and b, the process of filtering keeps you really engaged.

So that ultimately, it is your decision and

choice about what you decide to consider putting into your point of view.

And that kind of gets me to the final point, which is to integrate.

And filtering and integrate really go together.

So, you've got to decide what you're going to actually integrate into your

thinking, into your point of view, into your company strategy,

into your interview processes.

Ultimately, though, you need to be accountable for

what you choose to integrate into your thinking.

And I would say that part of accountability is to make sure that you've

reflected on it, that you've validated it, that you've tested it through debate,

and that you have chosen wisely the kinds of things that you want to integrate into

your thinking.

So, those are five key actions that will help you be a better thought partner and

get more out of the process, and also, I think,

help you avoid some of the pitfalls that could

be associated with relying too much on generative AI models as a thought partner.