Shared conversation

Claude-instant-100k Β·

create a blog post to accompany a video for which the transcript is below. Make sure it is wity and about 1000 words
πŸ“ helpline. How can I help you?

Good morning, Ricardo. How you doing?

Doing great. How are you, Ian?

Yeah. Great. Fantastic. Good to chat. We got loads of things we could do on any kind of our podcast systems, but I particularly was enjoying your productivity hacks and wanted to ask you to give me some help on something I'm doing and I think it would make a great episode.

Crossover episodes are always a good thing as well, aren't they? So I think make a great episode, but is that okay? Mate?

Yeah, that sounds good. Let's let's apply productivity hacks to a problem.

yeah. Yeah. So


there for.

Here's what it is. I've, I write books, I write Christian books. I want to take a, an audio, it's a sermon. I do, and I've got an audio and I've got a transcript of the audio already.

What I need to do is turn that into a 3000 word chapter in a book, and the book is a book called Built to Last. That's the name of the book. I've already got all the chapters I wanna write, but they're all sermons. That I started with and I wanna, write them. And now last time I did this, it took me about three months to do all the work of taking all the sermons, get the transcripts, edit them myself, bring it down, make the book Cohi coherent.

I've done that now, I've now got a very coherent book. My first book outta here, God. So this is the second in that series, and I've got a deadline and I can't take three months to write the

Yeah, let's try the 15 minute edition.

Yeah, go on man. Guide me through what I need to do.

This based on actually what I taught on the last episode of Productivity Hacks. So it's really cool. I know you already signed up for po I'm pretty sure cuz you watched the episode.

So please log into po.


β€Š πŸ“ Gimme a second. So on the screen so it looks like po to anyone who's listening, by the way, this is live. We haven't planned this. I dunno what Ricardo's gonna suggest I do and how to do it, but we'll go where it goes, which is cool. So let me put the code in. On the screen.

There we go.


Okay, here I am, Ian Banner in poe. So I can see already. So I have a chat, G p T four sort of subscription. $26 a month. I can see already this has got chat. G P T Claude for other things. Dragonfly new. You have to talk via those. So it looks like I've got a load of different chat box and AI systems to use.

yeah. And all on one subscription, which is really cool. So otherwise you would've to sign up for each of them and pay for each of them separately.

There are a few limitations. If you have shot Shipt four, you can talk and you pay open ai, you get unlimited shots with it. Here because they are paying for it on the back end of your subscription.

They give you a limitation, but the limitation is quite reasonable. They say, so you try clicking on GT four and you see it on it's unlimited access, but you get quite a few messages to it. If you click on GT four on the top, it should tell you what the limitation is.

It doesn't actually say, but if I remember when you're signing up it tells you can send it a thousand messages a month or something. So it's, it


a thousand a

cost effective way

by 30 is,

Now how long is your sermon? Was it like a Twitter kind of sermon

Ooh. No. No. The original sermon we're talking about was an hour.

Yeah. Four won't help you much.

Ah, yes. Actually I tried it in G P T four. I remember pointing in and it said too long.

So I said, oh, I'm gonna have to break it down and I'm gonna have to do a lot of the work myself. N nothing wrong with that, by the way. I don't mind doing the work. It's just I realized there was gonna be a bit of a process there that I, it seemed a barrier.

Why? Why do I need to cut it into smaller bits? Send it in, get transcripts, Reed, just seemed a lot really. So I haven't got on with it to be blunt.

As I said on the episode, there is a war brewing, which is the context war, and what is context

when you're talking to an AI to an l M. It remembers the last few things that were said.


And uses that as context to continue the conversation. For example, G PT four has a context of 2000 tokens.


you can think of a token as a word.

It's not exactly the same. It's like 0.5 75 tokens.


One word is like a token and a half roughly, so you can just think of them interchangeably. Now, Claude has launched two models. Claude is built by philanthropic, which is one of the competitors to open ai. And I explain on the episode how they differ and all of that, but they created two models.

One of them is Claude Plus, which is very similar to Chatty four. So it's really smart, but it also doesn't have much memory. Much

tokens. Tokens are like memory. It's like the conversation we've had.


At some point, I'm gonna get a bit of Alzheimer's and forget what we talked about yesterday.

And because this is a computer, you know exactly how soon it gets Alzheimer's. Now if you look at on the left, you should have a hint that there is one with a very limited Alzheimer's

Claude, instant foot hun. A thousand K. A hundred

a hundred k.

I thought that was the speed that, that's the context.

that's the context. So this one will remember the last a hundred thousand tokens. Just to get you some context, a thou 10 a hundred thousand tokens is a 350 page book or probably an eight hour sermon. Do you have yours? Do you have a transcript of your sermon?

I do. Yeah. Okay, so let me get that out.

yeah. Should just put it on your clipboard.

And let me show it you. This is this, I'll show you briefly. Here it is. This is the transcript. And by the way it's just, as you can see, it's just all it is. There's there's a some verses from the Bible there. And literally I'm chatting to people. As you can see. It starts with, I want to start to remind you why this letter from Paul is quoted above where we were yesterday, cuz this was at a conference.

I'm talking to the conference, it's unclear what words were being used at various points by the transcriptor.


get the gist

on the top right. You have a hint. This is five, 5,600 words. Solar, roughly 8,000 tokens.

I think in the future we'll see in the editors how many tokens this is, because this is the new currency, not the word. The token.

The token. But I can think of tokens as about one and a half words.

So this will be less than 10,000 words, less than 10,000 tokens.

Okay. Let's grab that then. That's in my


grab that, chatty four would tell you that it's too long, but you don't even need to try it now. You need to tell Claude what to do with what you're sending it. So I would say I will give it a bit of context.


You could say something like, I'm gonna give you the transcript, or I'm gonna give you below the transcript of a sermon I gave, and

and I am going to ask you questions about it.

Yeah. It is part enough to know that. But it's always good to be to you. Never error. Now don't press enter, press shift. Enter


and now paste your sermon.

there's my sermon and then I'm gonna hit the moving forward button. Yeah.


Okay. There we go.

It's sending


that's my scroll. That's everything I've gotta say. Wow. That's fast. That is fast. Ricardo.

is That's the other side of, that's the other side of Claude. It's 100 K and it's instant and it deserves it,

Now I would with, I get the idea that chat, g p t is being used by everybody and their dogs now. And I pay a subscription for that, but it still scrolls across the speed if I asked it that question and I couldn't put that in. And let me just see if it's done, if I was cameras or.

The good thing is it'll tell you what it understood about the sermon, so you can check, does these look like what I talked about? So it gives you a quick summary. If you feed it information,

it's even asking

actually I really like what it's summarized, so that tells me two things. One, probably the transcript was good enough and secondly it's picked the things that matter in that tra because that sermon by the way, just let you know that I've preached that sermon maybe 30 times. So I do know the sermon very well.

It's one I use when I travel. So that is yeah, that is really good actually.

yeah. So now what do you what? What do you want to do with this?

So it's gonna be a chapter in the book. So maybe I should say, please summarize this into a chapter structure,


For a 3000 word chapter.


Okay. There,

I would correct through tier. I think it's far away enough from structure that you might confuse even Claude.

You and Claude, I've realized that you and I are gonna have conversations with a third person going, what would Claude think?

Yes. Yeah.

A new a new a new writing partner for all podcast work is



Okay, so ready? Let's see. What does with that.

Yes. Does sound like a good structure.

Yeah, and actually it's gone back. This is really interesting. The structure. This is really interesting. I gotta comment on this. This is really clever because the structure is correct, but that doesn't match the volume of words and where the words are. What I mean by that is most of the words that it had were in this false secret to endurance and an attitude of gratitude and challenge, and seeing gifting as an investment.



The pleasing offering, abundant provision. But the structure is correct. It starts with, we are talking about enduring ministry over time and it finishes with going back to that. That's perfect. I like that.

Yeah. And one, one other thing I really like and that's po actually PO always tries to give you ideas about where to go next,

and that is po not Claude.


So Paul is the interface into the shot, into the l M, and Claude is the model, but regardless what model you use, Paul tries to give you ideas of where to go next.

But knowing you, I know that you know where to go next,

But you could keep on exploring the subject with him.

yeah, it's it's saying we can do a bit more on cultivate commitment to God, which would be brilliant. But I love that structure. I gotta say again, Let me just be clear to everyone chat this this is Claude. Claude isn't writing for me. It's editing my sermon and my transcript.

It's doing a job for me. It's not creating, just wanna make that clear. I'm not asking it to make up a sermon on endurance. I think he probably could do that very well, actually. But just to be clear to anyone who's listening, I am the author and copyright of the original content. I'm asking Claude to do some work on it.

Okay? Now,

if I say,

How long do you want? The chapter is the next question.

Let me give you some feedback. That's great. It's each chapter, so the book's got seven chapters cuz there's one per day, and each chapter is about 3000 words. That's great. Please. So I'm gonna assume a little bit of an edit might be required by me, so I'm gonna make it 3000.

The last book had about two and a half thousand words per chapter.


In fact, what unplugged in, sorry, lemme just give me a sec. That's what they look like

when they're

Ooh. Pretty

saying. I'm back. Okay, so that's great. Please write this as a 3000 word chapter. Take that to this. Now I have a slightly sort of informal guidance style of writing that I deliberately use, so

maybe I should say that as well.

You could also paste a chapter of a book he previously published because it has enough memory. But let's not do that and say, please write it in the same style as

Ah, okay.

published before, but for now, let's keep that.

by the way, that could be brilliant because obviously I have a completely other book that I wrote to a style that is me,

Yeah, you can

tell it to write to your

I could tell it here is a style of my writing. Please analyze this and ensure everything you do is this style. Let's try that then. So first of all, let's say please write this in the informal guidance style.


it'll start me cuz that's what I tend to say to chat g p t. Please use informal guidance again, let's go. But maybe I can insert a chapter afterwards and say, please analyze this and rewrite. Yeah.


See what we do with that.

And it is fast

Nice. Super fast.

and y

Is that really 3000 words?

maybe it didn't quite obey on the 3000 word thing.

Let's have a quick chat. Sorry, I'm just

You don't need to select like that by the way. You have a share button. You have three dots on the right


copy message

message. Ah, okay.

And if you go to sis

And start a new page, it copied in markdown format. So you leave and have all the. So if you do start a new one and then paste this.

Cool. Let me find yeah, writing will dear. Ba. Why can't I? I'm fr

Yeah. you?

have a button on the top

like, Yeah. New. Just paste what you copied.

That's only 490 words, so didn't get the tent. 30,000, 3000 word chapter.

Try, please expand to 3000 words. This is called the demo effect, but I found that chat c p t also does the same. You ask it to write 2000 words and.

Yes, please expand to 3000 words. Okay. Okay. That did it clearly. Yeah.

Let's see. Copy message, or maybe it is message from God that you need to write shorter chapters. it did what it definitely says, I'm sorry, Ian, you only have material for 566 words. Go. Give me some more material that I'll write you something longer.

Yeah, it's 566 now. Maybe I need to be a little bit more specific about how to do it. Maybe I need to say something like, please take each paragraph and double the size of each paragraph.

But I think we can leave that for another episode. How to get AI to actually write how many words you need.

Maybe it's because it's a number. Maybe I need to anyway. Yes. Okay. How are we doing for time? So we've gone 16 minutes. How long are your normal episodes? Remind me.


Five minutes.


I'll do a bit of editing.

has gone on a bit then.

This is No, but it's pretty good. It's pretty good. But, so let's just summarize what we have learned today. What did you learn today From from what we discussed?

There's more AI going on the chat. G P T for sure.

And I'm always aware of the fact that, the bandwagon. Yes the world has changed, AI has arrived. If you're a writer, if you're anything in the content area, the world is different and you need to know that. That's certainly something.

I like PO a lot and I like Claude Instant. It was very fast and it had a lot of capacity.

Yeah, exactly. And I'm pretty sure that because Claude Launch launched the 100 k opening, I will launch the. 4 32 it is already available on beta, by the way,

Yes. Yes, we know.

is, yeah. So what we can learn is that to do what we just did two, two weeks ago would require, Python, would require a graph database would require running something for an hour, which would insert your book.

As little tidbits of 500 tokens into a graph database and would require a programmer and would take you the whole afternoon, which would still be faster than the three weeks that before it would take to do manually. So context is really a big thing. And I think just like we watched the memory wars in the eighties and the gigahertz wars in the nineties.

And the number of cores, wars in the 2000 tens and the performance per what wars in the 2000, in, in the late 2000 tens. The next war is the context war. That's where models can differentiate until, of course, you have a million tokens context that no one cares


right. It becomes a

then we'll fi then it'll be again.

So it's a combination of how smart you are. Which clearly CLO plus if you test it, is a lot smarter than Claude 100 k with how much context you have. So the next step is obviously having something as smart as CLO Plus, and with as much memory, putting double coats around it as CLO instant if you try CLO plus.

You will see that it's also really fast. That's one thing I like about Clot Plus. It's as smart as chat T four, but it is really fast. So try asking it the question and you'll see.

What I could do really easily here is put a question in that I might put in to chat g p t. So let's do that live. I don't mind. So what you're about to see is me putting a question into chat, g p t about a book. I want to write something on a book, and I'm asking chat g p t to be a research assistant to give me some information to think about.

Just to be really clear to anyone who's listening, this will not become what I put out as a summary of the book. It's a research assistant giving me some things to think about, just so you know. So I put that in, that's exactly what I put into chat. G P T. How fast? Oh, it's very fast. That is definitely fast.


That is quite fast. Wow. Wow. Good. Super fast. That is really fast, Ricardo.



it also forced shat to improve because now, for example, in cha t, I don't know if you noticed, but now you can also copy as markdown. It's much better at continuing. So this competition between Po Shai pt, it's interesting that you are using two separate tools. I really like. That now you can pay one company, Quora, which is one of the key producers of information, cause they have one of the big databases that these

models were.

So you can give Quora some money, which is well deserved cause they've been contributing to the community so much.


And then they are paying open AI for the API calls that they use and they are paying cloth for the AI calls that they use. So the money that you give to them will be distributed.

By API call, depending on how much you use. So you'll be rewarding the model that gives you the best answers by giving that company the most money,


All brilliant.

good thing. It's a really good thing. I'm totally behind PO and if you install it on your mobile, you can continue the conversation on your mobile.

There's so much I could explore here. Like the idea you could create your own bot. I don't even know what that will be, but

That will be for another episode, but it's really cool because you can create your own bot. Another cool thing is if you have a learning, you can click share and you can share that learning with the PO community. So you had a learning here, so if you click share,

it's just above. Tell me more. On the right,

on the

That you have a and dislike and the share.

Oh yeah. Share. Yeah share.

So you'll see that you have a share link and you can share this conversation now. Open it on a private browser. Or not private but I think private will work best.

You have a

just because it'll get me outta the context.


Yeah, I understand that. Let me gotta get there. I don't usually do private browsing, so let me find it.

File new incognito window. So you could share this to someone random on the internet and share it as a tweet.

That's what yo go, yo. Perfect. Talk to Claude. Yeah lovely.

So it's a good way of sharing an insight that you got without having to do any kind of edit. So if you have an insight, I want to share this in the world, or we could give, in our podcasts, we make sure that we share the background research that we did so that people can see how did we arrive to these conclusions.

So you can have a, we can have a conversation with Paul, record it, and then share. This is one of our sources and people will be able to audit what we did, which I think is also quite powerful. So starts to become a bit more academic

Very good though. Brilliant stuff mate. I put a post out saying to people about your your new series. This guy has changed my life. This is why you've changed my life, mate, cuz you know this kind of stuff. And we talk it through. My next book is gonna, and it's, I just wanna emphasize this to anyone who might listen to this.

It's still my work, it's still my original content. Please don't misunderstand what I'm saying. My next book, I'm gonna get to public. Publish in weeks instead of months. For sure. For sure.

and that is why. We need to use these productivity tools. That's my stance. We also need to have very important conversations about the role of AI in society and all of that, but just ignoring that it's out there and not taking advantage of this beautiful gift that human knowledge gave us, that is just stupid.

It's gonna change what we spend our time doing. Not stopping us doing things. For example, just briefly, I had a conversation with a university lecturer last night at a meal and he said, how are we ever gonna certify people's work at university if they, you've got these tools? And I said, by interviewing them, by talking to them, by having face-to-face conversations to know they know this stuff.

And he literally said, but that's gonna be really hard work.


And I went, yes.

Yes, that's what you'll pay the big bucks for. You don't need to regurgitate facts and create exams anymore. You need to have Powerful conversations with your


You need to go from teacher to coach.

Exactly. Exactly.

write it in the third person

Here is the blog post rewritten in the third person:

How AI Can Help Authors Write Books Faster

Transcribing long audio sermons and turning them into book chapters is a time-consuming process. It took Ian 3 months to do this for his first book. But with the help of AI tools, he was able to significantly reduce that time.

In the transcript, Ian speaks with Ricardo from Productivity Hacks AI about using AI tools to extract a 1 hour sermon transcript and turn it into a 3000 word chapter for Ian's next book in minutes, not months.

They experiment with Claude Instant, an AI model from Anthropic with a massive context of 100,000 tokens (equivalent to 350 pages!) that allows it to summarize long form text. After feeding the full transcript, Claude Instant provides a surprisingly accurate chapter structure and even suggests further points Ian could explore.

However, the initial summary is only around 500 words. Ricardo explains that to get AI models to hit an exact word count requires being more specific and even asking it to expand each paragraph.

The key takeaway is that context is becoming increasingly important for AI models. Models with more contextual memory like Claude Instant are able to perform complex summarization tasks that previously required a lot of manual work.

While AI won't replace human creativity, when used judiciously it can help amplify our efforts. In Ian's case, it will allow him to publish his next book weeks faster by automating some of the repetitious transcription and summarization work.

However, they must also have important discussions about the appropriate use of AI. While these tools can boost productivity, we need to ensure they don't replace critical human functions like evaluating students' deep understanding.

In the end, AI augments - it doesn't replace. The responsibility is still on us, the human creators, to generate original content and have meaningful conversations that machines cannot. AI is a tool to help us spend our time where it matters most.

How might AI change the way you work? Let me know in the comments.

This conversation may have been edited by the sharer.

Claude-instant-100kTalk to Claude-instant-100k on Poe