Lemme introduce myself.

My name is Alistair Matchett.

I, my start of my career in banking.

I started out at JP Morgan doing oil and gas and financial institutions m and a, then went to private equity at a fund called three I in London and then went into the training business.

And we have been teaching across Wall Street and London, actually around the world, um, finance for many years, dare I say.

So today we have a pretty short session of 30 minutes and what I want to do is give you some useful tools of how to use chat GPT because it's one area that I think will be transformative in helping you work.

So not just creating content, but actually helping you be more efficient and more effective. And that's really our focus today, is getting you to understand how you can use the system.

So what I'm gonna start with is I'm just gonna go to some slides I have put together and I will share some of the slides with you, um, today.

Um, and what, when I share them, they'll go into the chat box.

Now if you have any questions, please use the q and a box.

Okay? So if you have any questions, please use the q and a box.

So what I'm gonna do now is I'm gonna share my screen and I'm just gonna go there and you should be able to see my PowerPoint screen and let me just give, get my other little boxes popping up so I can see them.

So if you um, have any questions, you can just use the q and a chat.

I will put the, um, the file that, the PowerPoint file that I'll use into the chat as well.

Okay. So I will do that as well.

So, um, that's at the end of the session. I will do that.

Okay, I'm just gonna arrange my screen a little bit.

Okay, so let's get started about chat.

GPT prompting, and this is actually quite a sophisticated issue when you get into general GPT kind of AI machine learning, there's a real skill to prompting.

And what I want to do here is give you the kind of the basics of how to make your job easier, whether you are in investment management or whether you are in investment banking.

So we're going to start by just writing text, um, then extracting information from text. There's some really neat stuff that you can do to extract information and that could be headings, it could be numbers outta text.

That usually would take you an awful lot of manual time to do, but it's really, really quick with ai.

Then rearranging text.

You can take a document and you can move things around in the document, particularly for formatting, um, VBA coding.

So that helps you with modeling and then just show you how can you can learn things.

And then a bit of brainstorming at the end.

Now, before we just dive in, it's worth just understanding what chat GPT is.

Um, it's based off a large language model. What is that? That is just basically a massive amount of text that the system has been trained on.

Now the critical thing for chat GPT, it has a cutoff of September, 2021.

So this means you're not going to get from its kind of existing knowledge any information beyond that date.

So this means that if you are asking it something cold, it won't come up with up to date information.

However, what you can do is you can copy paste information from financial statements and filings into chat GPT that is current and ask it to analyze that.

So even though it's kind of memory stops at September 22, 21, actually you can use more up-to-date information.

Now the system is structured as a neural network and that's start out as life as a kind of proxy for the human brain where you've got lots of kind of small decision makers taking in data, processing it and saying, I'm gonna pass it on or I'm not gonna pass it on.

And the important thing to understand is that the results you're gonna get are probabilistic and not, what that means is, is it's not linear, it's not going to kind of, kind of follow a path down to a specific answer.

What it will do is it will kind of estimate what it thinks the answer is going to be.

Now, usually that's pretty good 'cause it's actually fairly good at getting answers, but sometimes it does get it wrong.

So you need to be careful with the system and it's more reliable if you kind of give it something to analyze rather than you let it coming up with stuff itself.

So I think, and from a finance context, it's safer to copy text and put it into the system than it is just kind of relying it, relying on it to come up with information itself.

But there are still really good use cases and that's what I'm gonna focus on this session is kind of the different use cases that you can get.

Now the first thing we can do is when you're writing text is just to understand um, the spec, the specific this to specify the text. Lemme come and explain what this means is if you just ask it, write a summary of what Kellogg company does, for example, it's going to do a pretty basic job.

It was much better if you say that you are a consumer research analyst and you want to write a summary of Kellogg for the investor community.

So what you're doing there is you are suggesting to the model your level of expertise of the, of the writing, and secondly, you are specifying who the audience is as well.

So those two things are pretty important because otherwise you can get some pretty basic results. Whereas if you specify that you are an expert in the topic, you're gonna get much, much better results.

And specifying an audience also will tweak the result for that particular audience.

The second thing is tone.

You can change how the, how the um, output sounds.

So you can sound informational, you can be lighthearted, you can be funny, you can be persuasive.

Just be a little careful with this.

I've been using it and one of my colleagues said, I know this is not you writing this email because it just doesn't sound like you.

So just be a little bit careful about some of the tonality because it can make you sound, um, alien or just not your personality.

So just be slightly careful about that.

Um, the, the third thing is length.

It's really important to specify how long you want because the language model will, um, be quite verbose or it can be quite short.

And if you specify the length and words or characters, you're going to get the result that you actually want.

And that's pretty important. In terms of how much text you can input into this system, it's about a thousand to 2000 words.

You've got about approximately about 4,000 words of input and output.

So what that means is you can input 2000, you may get out 2000, but if you input 3000, you won't be able to get out more than a thousand.

So there are limits in chat GBT about how much input and output you get and they're measured in tokens.

And so it's not exactly words, it's tokens, but broadly speaking, as a rough rule of thumb, I think 4,000 words, combining what you've inputted and what you've outputted is kind of what you're gonna get.

And you'll see that if you put too much text in, it will just say it, tell you you've, you've got an error.

And then lastly, in terms of specificity, specificity is just the emphasis on areas.

So you may want to say, I want you to focus on Kellogg's products or I want you to focus on Kellogg's segments and it will focus on those particular areas.

The second major issue is that when you are typing in the prompt is the way um, it's structured. And let me just jump to the the page 'cause it's probably worth actually doing that.

Now if I go to chat BGBT, and I'm gonna go to my Kellogg's segment analysis here.

So you can see here fact if I go to um, my, oh, I thought I had, here we go. 4.1. So this is one, I've just done this earlier because it takes a little time to get the results.

And so I don't want you just to be sit there watching results.

So here it says, write a company description of Kellogg.

There's no tonality, there's no who's writing it, there's no who's writing it to.

So it gives actually a pretty verbose overview of Kellogg.

Then I've said rewrite this in 50 words, it's much tighter.

Um, so that's helps that.

Now if I go to my other example here, what I've done is I've been much more specific.

I've said you are a consumer equity research analyst.

So what that's saying is that you are an expert, right? Common description in 70 words, that's the length.

Um, Alan mentioning its key brands and focusing on the size of different segments. So I'm told what to to specify in the text and then tonality, I've said write in a tone to persuade someone to invest in Kellogg and start with an introductory sentence.

Now that's the other thing I wanted to mention is that you not only can you iterate and build on the prompts and you can see the way it's organized, each prompt you do a new chat and that will be a set of prompts and it will remember what you've prompted within that chat, but it won't remember between chats.

Um, you can also specify the format.

And I find this particularly helpful because you can say, give me bullets, give me a table, put it into HTML or put it into markdown.

And markdown is a little bit like HTML except it won't give you the actual code but it will be formatted.

And the nice thing about that, you can just copy and paste it into Excel so you have some kind of pre-formatted documents.

So let's go to my first case use case, which is company profiling.

Remember the cutoff date is 2021, so it's probably better to copy and paste content into it if you want to do this from the company's 10 Ks business profile.

And I've done both, but in the first instance I'm just asking it to write content.

Um, so here I've said your consumer equity research analyst, that's credibility.

Um, 70 words what to focus on what Tony want.

And then I've said exactly what format I want.

I start with an introductory sentence and then put the rest of the content into bullet points and the segment breakdown into a table.

Include the sources. Now I found sourcing in chat GPT really unreliable 'cause often the links don't even work.

So just be super careful with that.

I found some of the other tools better in sourcing than chat GPT.

Okay. And I've also found it pretty bad for m and a acquisitions, but if I just go back to chat GBT, um, you can see here that um, it's saying that it can't gimme information beyond 2021, which is fine.

But then I've got this section here and what you can do is I can just copy this code, I can open the notepad here and I'm gonna paste it into the notepad and then save this and I'll save it on my desktop and I'll call it um, Kellogg three because I've done this before.

Um, and you can save it as h TM l you may be able to say actually save it as a um, word doc.

Yeah, I'm just gonna save it as that.

And then what I can do is if I open um, word, I can then just go and open that and I'm gonna go to my desktop.

There we go. And I'm gonna open Kellogg three and it pulls it into Word and you can see it's got a title, got some bullet points and it's got a table here. There's a bit of formatting you need to do, but at least it gets you further along the track than it would be if you were just copying and pasting having to reformat.

So that's just helping you be more efficient.

Okay, the um, second case profile I've got is where you'll, you'll get more reliability because what you're doing is you are copying information from a filing into chat GPT and saying use this data for the output.

And my experience of this is you get better accuracy and also you get more UpToDate information as well because you are using current filings rather than just whatever chat GPT is going going to give you.

So I've copied the business script from latest 10 K filing and then you can normally expect an output of between a thousand, 2000 words and you'll have a limit of about 2000 words going into the system, two or 3000 words.

So just be careful about how much text you can put into the system.

So in this case I've said you're an investment banking analyst course, produce a company one page summary using the following information.

So let me show you the prompting that I did here.

If I go to number five, so in this case I go up to the top, um, you invest bank analyst, which is one page summary using the following information.

And this is just the business description from the 10 K for the company.

So if I just go to Kellogg, I literally went into the business description and I think it's in the latest 10 K.

Let me have a quick look. Um, I think it's, there we go.

So I literally just copied this section of the 10 K into chat GPT and told it to give me a company profile.

Now what it's done quite nicely here, I didn't ask it to do this but it's given me the output in markup language.

So you can see here it's got the headings, um, it's got overview key brands.

Um, and you may want to kind of just, just adjust this.

Let me show you another example.

Um, and this is something that I'm gonna go from the top.

It's got some other little features in here that you'll probably find useful.

So summarize the following text into bullet points.

And this is an a section of apple's 10 K.

So let me just go down and see what it's done. So it's company background and it's given a list of products and a list of services for Apple.

So this is just copied out of the 10 k filing a lot of information and it summarized it.

Then I told it to translate into Mandarin for a bit of fun.

I'm gonna talk about the translation in more later.

Um, and then um, translate to Hindi.

And then what I've done is put Apple's products into a table so I can copy it into Excel.

So again, this is stuff that could take you a lot of time manually to do this, but let me show you if you want to copy this into Excel, I'm just gonna copy this.

Control c I'm gonna jump to Excel, I'm gonna just do a new file there.

Now if I just paste it in, you are going to actually, that hasn't worked too badly, but if you do edit pay special and you do HTML often, it doesn't come in very well with HTML.

So what I normally do is just type it in this text and it comes in in a kind of unformatted way.

So normally just copy it, paste specialist text and then you can manipulate it and format it.

But that's a, a really nice little thing that you can do out of um, tables. And then I've asked it to rewrite the description in a persuasive tone, almost like a kind of selling document.

Okay? So if you are going to be using chat GPT for company profiles, I personally would copy and paste information into the system.

Now, one area I love and it literally has transformed my life in certain circumstances is extracting information.

Now remember there are character limits about 2000 words, two 3000 words you can put into the system.

So you want to specify the format of the output.

So in this case I've got a example using the information below, prepare a bulleted list of the main sections.

So if I go to the example here, getting to my number six.

So what I've done here, lemme just show you because this, I've just, I love this.

I went to the credit analysis playlist and literally did this and I've selected the whole page and it includes graphics, includes notes, it includes titles.

So I copied that and I've said using information below, prepare a bulleted list of the main sections and you can see this is all the garbage that I've copied and pasted in and it's given me just the main sections in a bulleted list. And this is because the client said, okay, what do you have on credit analysis? And initially bam, it was so quick to do.

So chat g BT is very, very good at just sorting out text to make your lives easier.

Let me give you another example.

Um, in this case what we've got is the list of um, webinars that we've got here.

So this is the webinar list and I think I've got this open. There we go. This is the Felix Live webinar series.

And what I wanted to do is just pull out the titles in a bulleted list.

And you can see here what's happened is it's taken all this text and it didn't do a very good job there.

And I've tried again and it's given me a nice, um, little analysis here.

So this is very, very helpful.

It saves a lot of that really kind of grunt formatting that you have to do.

So let me go to back to my um, prompting or look back to my slides for a second.

So that's extracting information.

But what you can also do, and this blew my mind when I found out about it, is you can rearrange text Now this is really, really good and if you've ever been given a document at work where someone says, oh, the date should be at the end of the line, not the beginning of the line or vice versa, and you just, your heart sinks 'cause you know that this is gonna be a real time sync manually.

So here what I've said is take the information below and I'll show you the prompt.

Um, change the position of the date from the end of the heading to the beginning of the heading.

Remove the brackets, put a dash between the date and the heading and a dash between the date and the length of the session.

So if I go to my Felix live webinar, so you can see this is the text that I gave it, this is the heading.

And what I wanted to do is remove the brackets around the date.

I want to shift the date to the beginning of the heading.

I then want to put a dash between the date a, the heading and I want to put a dash between the heading and the 60 minutes.

And I want to do that for all the headings in the document.

Now you can imagine if someone asked you to do this, you'd be like, like this, just drive me crazy.

But Nirvana is here.

If I go to chat GPT and I um, go to seven.

So you can see here, um, at the top what I've said, take the information below, change the position as I've got on the slide and I've given it all the live webinar series content.

And then let's take a look at the output.

So this is the output you see, you can see this is just, just makes my heart sing.

It gives you the date, it gives you a dash, it gives you the heading dash and then the minutes.

I mean how good is that? Um, so I've just had a question.

Can chat GBT be used as a web scraper? For example, if I want to extract the 10 K file from a Lego database by sim simply bypassing the URL, um, not at the moment.

Um, you can't do at the moment I have tried this because, and we actually trying to put this into our Felix platform exactly that, but it doesn't do that at the moment.

It's not very good at using current data.

No doubt that's gonna come in.

And I know there are tools out there at the moment, but it will eventually come in.

But at the moment you can't use live URLs from Edgar to get the information unfortunately.

Okay, let's continue.

So I have done the um, rearranging text and that really for me is just like, just blew my mind.

Now the other thing you can do is if you're doing comparables, um, you can get it to list companies.

And this actually really helpful in unusual sectors.

Bear in mind you've got a cutoff date for September, 2021.

Um, you may want to say how many companies you want, particularly if there are lots of communists in the sector.

Uh, you can specify gimme a ticker, you can also give a one line description and you can ask to have results in the table.

I've done this full transaction comps and it's pretty crappy for transaction comps.

So I probably wouldn't w worry about transaction comps.

I would use it for trading comps, particularly in unusual sectors.

So in this case I've said your investment banking analyst, here you are list five companies in the specialized animal health industry in Europe.

So I've said to the geography, geography provide one sentence for each company and format the result on the table so I can copy this to excel where possible use the ticker and an estimate of sales.

Now bear in mind because of the 2021 cutoff, the sales are gonna be out of date.

But let's take a look at what it produced here.

So this is the company ticker description and this is pretty good start, right for, I don't know anything I did.

I mean the reason I put this in, I did do a project in the animal health industry and it was just completely um, mind numbingly dull and I didn't know anything about the industry when I started.

So you can see here I um, have got a pretty nice List here of companies in a very specialist sector and if I was doing comps, this would be a good place to start.

Um, financial modeling.

Now I have had, you need to be a bit careful here because the, the best case use really for financial modeling is building macros and VBA macros.

And um, remember you can't add do macros. So make sure that you save the file before you use this.

And here I'm saying you're a financial modeler.

Write a macro to turn all constants or hard numbers blue.

Um, it's to say in an Excel workbook.

Also make the font 10 point aerial the zoom protection a hundred percent every sheet and put the cursor on the top left hand corner on every sheet.

So let me give you an example of the file I wanted to clean up.

So this is some analysis I did pretty quickly.

So it's got um, different sheets.

Not all hard numbers are in blue, it's different zoom levels is in, it's in um, different fonts I think in some places as well.

So let me just show you.

Um, if I go to here and I go to the prompt I put in, what I did is um, I put in a Euro financial modeler and write a macro to turn all constants or hard numbers blue, um, in an Excel workbook, um, make the AEL 10 point font and the projection a hundred percent.

Now it's giving me VB um, VBA code here and you can see it when it does it and you can see here.

Then it tells me actually how to input this.

Then what I'm going to do is copy this code, I'm gonna go to Excel.

Now you have to make sure that you save the file as a macro enabled workbook.

So I'm just gonna check that I've done that.

Yep, I have done that. So I'm just gonna save it.

Um, and then what I'm going to do is you need to make sure that you have the developer ribbon turned on.

So I'm gonna go to the developer ribbon, I'm going to go to Visual Basic.

You can do alt F 11 to open that.

Um, and then you can create a new module, insert a module, paste that code in, save the file and then you can run the code.

So I'm just gonna click run and you can see it.

I go back to the model, it's put gone to every single sheet.

It's put the hard numbers in blue, it's changed the zoom, it's made sure the font is aerial, just move through the file and it's done a pretty nice cleanup.

So that's a pretty good case of where a macro can save you a bit of time with formatting.

Okay, so that's a financial modeling use case.

You can also, and then we're nearly out of time, just a few things more.

You can also get it to answer questions.

So in this case calculate cash open financing activities.

And this is from our, one of our quiz questions.

So this is a little dangerous in that obviously people can cheat, can cheat in exams.

But here if I go up to the very top, um, I put in the question and the nice thing about this, and this is really helpful if people are learning, it doesn't actually explain what's going on, it doesn't just give you the answer, it actually explains how it's calculating the answer.

So you um, can actually learn a lot.

Now if you want to restrict cheating, cheating, what I um, recommend you do is you use graphics rather than text.

Okay? So it's important to use graphics rather than text so people can't copy and paste questions.

But I find it really, really helpful at learning if you are learning how to do something.

Translation. Translation is super good. This is probably one of the best cases that you can use.

And let me show you what I did with translation.

This is Adidas is Financial Statements.

Now they do have an English version, however many private companies don't have English versions.

So you get a private company set of accounts.

Sometimes you're looking at this and I don't speak German and it's kind of just complete gobbledygook to me.

But what's painful is that if you had to translate these individual labels without having the whole table.

Now what I've done here is I've asked chat GPT to translate the information and obviously quite a lot of information translate into English and put into a table and it's done a really pretty nice job down at the bottom and you can copy this table into Excel.

There we go. So this is that filing converted to English and put into a table.

So it's a pretty good start if you've got a private company.

I've done the same thing with a little Mandarin example.

So this is just press release and of a company called Mind Gym and I've just asked it to be translated into Mandarin and it's really, this is, I think this is probably one of the areas where it really excels is translation.

Um, and then you can do brainstorming and frankly you it's actually pretty useful creating up ideas or titles.

Um, you can do limericks and your poems, you rhymes, you can do things that will have the first letter.

All the same you's actually pretty good at linguistic gymnastics.

So um, that's a bit of fun.

But we are kind of at time Right now.

I hope this has been useful.

I'm starting to use chat GBT every single day.

Um, I think at the moment it's the best product out there.

There's bared at Google, which I've used. There's an add in to um, Microsoft Edge, which I use.

So I actually use this a lot.

Bear in mind it does use your data, you can switch that off.

But if you switch off not using um, not allowing it to learn from your data, you won't get a history of the chats so you can turn that off so it doesn't learn from your data and that's obviously particularly relevant if you is any confidential information.

And so I would not put confidential information into the system at the current time, but um, I really, really appreciate your time.

What I'm going to do now is I'm gonna put um, the, if I just open that's my little, I've got so much open, lemme just open a Windows Explorer.

Um, I'm gonna put into the chat a um, version of my slides.

If you just gimme a moment while I find them. Here we go.

I'm just gonna go and get my slides 'cause I did um, put them in and I'm gonna put this in the webinar chat.

Okay. I'm just putting it in the webinar chat so you should be able to see it in the webinar chat.

Not the question and answer but the webinar chat, these are the slides that I've gone through, okay? So it kind of gives you just an overview of what I've covered and the suggestions in prompting.

But I really hope this is useful.

The session is recorded and you will see it in a playlist coming up soon.

But um, I hope you have fun with chat g PT and let us know if you have any questions.

Um, you can ask through Felix through the instructor panel.

But I hope you have a great day. Thanks very much indeed for listening.