🎙️ Unlocking Daily Podcasts in Just 2 Hours with ChatGPT 🎧

Hi there

Today, I made the newsletter a practical one (no pun intended).

Less memes and more action items.

But if you don’t care about podcasts or use cases for ChatGPT, then today’s edition might not be the one for you…

But, but… you love memes?

Okay, if you came for the memes, here is one just for you…

Using ChatGPT for rapid Podcast scripting and creation

In August, I am running an experiment to see if producing a daily podcast in one hour a day is possible.

I want to find out if I can run a daily podcast without any external help and without it taking up so much of my time that it affects my other businesses and projects (of which there are a few).

Some would say, what’s the point of wasting all this time on a podcast…

A podcast can serve multiple purposes.

For some, it is a way to establish connections with other professionals (interview podcasts); for others, it is a way to build thought leadership.

You may have another reason why you might want to have a podcast.

In this experiment, I am thinking about a podcast as a way to build thought leadership and an audience - or, in another way, try to get known as somewhat of an expert in the field of AI.

Now, in this podcast specifically, I am not really sharing any deep thoughts or wisdom. I am basically regurgitation the news.

It is an experiment, after all…

But my thought process is this:

If you have a podcast or want to do a podcast on something you are knowledgeable about, you can take my process and expand on it with your own commentary and wisdom, and that way, make it more valuable.

What I will share with you today is not a fully automated process. It will require some work on your part.

AI will make you more effective, but this does not replace (all) manual work.

AI is like Redbull, “it gives you wings.”

Without any further mambo jumbo, let’s get to the practical stuff!

How I produce my daily podcast, AI at Dawn

First, let me clarify that I am using ChatGPT GPT-4 with the code interpreter for the AI part.

I also subscribe to a service called Mailbrew. Every morning, I get an overview of articles that were published in the last 24 hours on my topic of choice (AI in this case).

In addition, I subscribe to various newsletters like Reuters, WSJ, Barrons, etc., and I check Google News.

Step 1
Every morning, I find five news stories that I think are important and could be a good fit for that day’s podcast.

For example, today, I did a shorter version, as I have four topics related to AI and (global) politics, and decided it did not make sense to mix in one story about advancements in cancer research with AI (I leave that for Monday).

Here is an example of a story I covered today and will use as an example here:

https://www.whitehouse.gov/briefing-room/presidential-actions/2023/08/09/executive-order-on-addressing-united-states-investments-in-certain-national-security-technologies-and-products-in-countries-of-concern/ 

I will find the stories and open each story in a separate tab in my browser.

Step 2
Copy the text from the webpage into a Word document (ChatGPT doesn’t like PDFs, it turns out). One article = one document.

Step 3
Once I have my five documents, I go to ChatGPT and give it my “magic prompt” (below), and using the code interpreter - which now can do multiple documents as of last week - I attach my five documents.

As you can see below, I ask for several things in the prompt.

Sometimes, the outputs for the socials can be used as is, but most of the time, it forms the basis for a Tweet or a LinkedIn post I must edit and create myself.

Here is my prompt:

Act as a copywriter and a script writer for my podcast on AI news daily. But pay attention to the facts and details. This is a news podcast. I do not want it to be chatty.

I do not want a “welcome” or “hello” intro or an outro.

Attached are some documents that I need you to do the following with:

1. Write a short summary.

2. Write a script for a segment on my podcast on this story. It should be no more than a 90-120-second read. The tone of voice of the podcast script should be factual yet not dry. It should have a mix of short and medium-length sentences. Create the script for an audience that is not advanced, so the concept should be explained to non-professionals. Easy and non-technical.

3. Write one tweet to entice curiosity.

4. Write one tweet that is provocative and polarizing. I want to invoke strong emotions in the reader.

5. Summarize the article's content into a 150-word post for LinkedIn. I don’t want any hashtags.

I want you to perform these 5 steps with each article I provide you.

You may notice that I am giving the AI specific instructions on what I do not want. I do this because it kept giving me all kinds of intros and funky greetings, and I don’t want that. I create my own…

Also, when ChatGPT allowed multiple file attachments, it would either prompt me what document I wanted it to work on or do one document first and then ask me if it should continue with the next. Hence, the final instruction.

Sometimes ChatGPT gives me too short scripts for the podcast, so I need to give it a “kick in the butt” with this prompt:

The scripts are too short. I can go through these in less than 120 seconds. Update the podcast scripts, keeping a balance between “brevity” and “detail”.

The important concepts here are “brevity and detail.”

Here is a link to the output ChatGPT gave me on the aforementioned story from the White House.

Step 4
I take all the podcast scripts into my “script” document (Google Docs).

I then write the intro (which is almost the same every day), and if will usually add some “transition” words between the stories.

Then I add the end.

Curious about what the scripts look like? Here are two examples:

1. AI at Dawn script 9 August 2023

2. AI at Dawn script 11 August 2023

Step 5
Once the script is made, I take out my fancy microphone, start Descript, my recording software, and start recording.

Descript is free for 60 minutes of transcription, then it costs $20/month. But it is hands down the easiest way to produce a podcast yourself!

[Descript is freakingly amazing - one of the best software out there!]

All you have to do is remove the wrong words you say/speak from the transcription, and it magically disappears from your sound file… It is so easy even my granny of 95 can edit podcasts…

I will need anywhere from 13 to 22 minutes to record - depending on how difficult words are in the script… If you record in your native language, this might go considerably faster…

As a non-native English speaker, I sometimes struggle with words 😂 

Some of the worst are “manufacturer”, “ubiquitous”, ”ambiguity,” and a few more…

Step 5
After I am done with the edit, I download the file and keep it orderly and nice in my Dropbox.

Step 6
I upload my podcast to my podcast host Podcast.co, add the descriptions, and press publish.

It is an amazing podcast hosting platform. So simple and easy to use. I actually bought a lifetime deal with them on AppSumo some years ago, and my gosh, was that a good deal… I think I paid like $49 dollars for three podcasts with free hosting forever! (There is no such deal there now).

But even their paid plans are okay, I think... However, since I have a free plan for life, I have not really researched what options are out there (or what are normal prices).

But Podcast.co is super duper easy to use, so I am very happy with them and can recommend them (not paid to say so, just a happy customer).

Step 7
Tweet something on Twitter, and post something on LinkedIn.

I try to make one tweet per day with this, but I do not post on LinkedIn every day.

I should, but I currently don’t.

Here is an example of a Tweet I did with the help of ChatGPT (but still edited manually by me).

I am not a Twitter guru, and you can see my engagement is not amazing.

One day, very soon, I will start taking both Twitter and LinkedIn seriously! It is a good business practice that I just have not gotten properly around to yet…

Other Notes
I used Claude 2 for the first two episodes. It was because of Claude 2 and its ability to read five concurrent files of any format like Word, PDFs and more I “came up” with the idea that I could do a daily podcast quickly.

In my brilliance, I thought, “What if I just find the news story, print to pdf, upload five articles to Claude 2, and ask for a script?”

It will take me 5 minutes or less…

Turns out, it is not thaaaat easy.

Sure, Claude 2 is better at reading PDFs than ChatGPT, but even Claude 2 struggles with PDFs, especially when all the images and ads get included in the PDF.

Hence, the text must be lifted from the webpage and put into a doc file.

I am sure there are ways to quickly scrape the various web pages and automatically get the text into the doc, but as this is just an experiment, I will not spend much time figuring out the automation. At least not for now… But I suspect BardeenAI will be able to help me with this, should I decide to continue after August.

The good thing with Claude 2 is that it is free, so you can try things out for yourself using that AI. Maybe you get better output than me, or output good enough for what you need?

Maybe you need a quick summary, for example?

Alternatively, you can also use the free ChatGPT 3.5. The only difference is that you then need to copy/paste the text into the prompt (you do not have access to the code interpreter in the free version).

GPT-4 is a superior LLM, but I assume GPT-3.5 is good enough for summaries and crafting scripts and tweets.

PS!
I used ChatGPT to come up with the name of the podcast, too 😀

Status after 13 episodes

I thought it could be interesting for you to hear how things are going…

My first 2-3 episodes took me 4 to 5 hours to produce. And if you listen to the first episodes, I even sound like an AI…

Now that I have the process down, AND I am getting better with the microphone AND better at reading a script, I usually can produce an episode in about 2 hours.

I set out to see if I could manage this in 1 hour or less.

I cannot imagine I will be able to do it in 1 hour unless I figure out some automation, like the aforementioned BardeenAI.

But I think it is possible to build an automation that does the following:

1. Give the AI automation my links to the articles

2. The AI (Bardeen) collects that information from the webpage and puts it into a Google Doc.

3. Then takes the doc and sends it to ChatGPT via their API. Runs my prompt, takes the output from ChatGPT, and puts it into another Google Doc.

4. Notify me when it is done.

Then, all I need to do is add it in a final script and do the recording, edit, and upload.

The thing that takes the most time is currently the editing. Editing takes time because I need upwards of 20 min to get through the script: longer recordings = longer editing time.

As I get better at reading the script without mistakes and stupid words like “manufacturer” (notice I never talk about “chips manufacturers” in the podcast, only “chips” or “chips producers”?), I think I will be able to shave ½ hour off the editing process alone.

Come to think of it… This automation stuff would be a cool experiment to try…

Damn it… Enough experiments already, Thomas!

My biggest pet peeve

One of my biggest pet peeves is poor sound quality on a podcast.

I disagree with the people saying, “Just start with what you got”.

NO!

Get a proper microphone!

It will elevate things a million times, increasing your chances of success exponentially…

I have a bit of a fancy microphone, the Shure MV7 USB mic. It is so easy to use and so insanely good!

Get yourself a dynamic microphone as it will “cancel” out a lot of room noise, and you do not need to care so much about soundproofing and noise outside your window or office.

But my mic is not cheap. There are other dynamic microphone options out there, but make sure you get a USB, especially if you are a beginner or don’t want to fiddle with a bunch of amplifiers, cloud lifters, and more.

Here is an article you can read if you are on the hunt for a microphone.

If you are serious about a podcast, do yourself a favor and get the Shure MV7 USB microphone. I promise you, you will never need another microphone, and you will love this one! (PS! You will need some sort of holder, too - just get a proper boom arm, not one of those $20 ones!)

That’s all I have for you today.

And while it was a bit different than usual, hopefully, you found it interesting and useful.

If you have any questions about AI or any feedback, just hit reply or tweet me @thomassorheim 

PS!
What's a podcaster's favorite type of shoe?

.

.

Anything that's "on-air"!

(Today’s dad joke courtesy of GPT-4)