Just Products #6: Using Elevenreader, NotebookLM, Wayback Machine, Make.com, buffer, ChatGPT and looking at warpcast & bluesky
Just Product brings you by-weekly Product Ideas, Stories, and Thoughts about Product Development.
Headlines this week
Eleven labs with their NotebookLM rival.
The Way Back Machine
What we created with a bit of Make & Buffer Magic.
The Reason for the Setup
The Solution
Learnings
Need help in creating something similar?
Eleven labs with their NotebookLM rival
I’ve been using Google’s NotebookLM product to create audio content here on Just Thoughts. You can generate a podcast episode from one or multiple sources, get summaries and briefing docs of the content, and interact with it by asking specific questions.
As someone who primarily always writes, being able to turn your writing into different types of audio content is the type of AI use case that highlights the technology utility value at its best: using AI to augment your abilities and extrapolating your ability to express yourself.
Now, the company that’s chosen to build open source and focused on AI’s abilities to translate content has released a product that supposedly rivals NotebookLM. I was first inspired to try their app when listening to their founders appearing at this year’s Slush event.
Eleven Labs app experience is relatively seamless in onboarding and feels intuitive when getting started. You first choose between 100+ voices with which you can listen to written content by simply copy-pasting the link into the app. You can try it with any of the Just Thoughts articles if you think it’s been too tedious to read all the content produced ;)
When trying the newly released feature “Make this content into a podcast episode,” it’s a one-click exercise that does a lot with just one click. Unlike NotebookLM, you can’t configure the episode in any way or choose the voices, which is a core strength for Elevenlabs products.
It creates an episode that adds content outside the scope of the article, such as “expert advice” and other things outside the scope of the written text in the original content. This may be a desired behavior when it comes to learning content, but it's not desirable for someone who wants to augment their content or use AI to create an audio version of their specific content.
For example, if I wish to do a podcast episode about the content in this post, I’d still use NotebookLM because it primarily covers what I want it to discuss. However, if you, as a listener, would like to a) listen to the written content as it is written or b) get a podcast episode about the general topics discussed here with additional perspectives, you’d want to use the Eleven Labs app. I haven’t found a guide for using elven-labs, but here’s one on NotebookLM shared by
on AI supremacy.
Nonetheless, the moral dilemma with the text-to-speech augmentation is how the original creators may or may not represented in the augmentation. In the Just Thoughts case, will you listen to this in the voice of a white-male-caucasian in native American, which would be the closest to what my voice would be identified with? or will you choose something completely different just because you like it better? What about content created by minority representatives in your community?
As consumers, the very least we can do is be conscious of this fact and use AI to mitigate bias, not extrapolate it. As creators of the technology, we should make the options available and inform users of what is popular and what is not, educating them on the implications of these trends.
We could even ask the user if they want to listen to the content in their preferences or respect the cultural heritage of the content creator, given that we can first use technology to identify such factors quickly.
The Way Back Machine
Without sounding like the Wikipedia page describing the Wayback Machine, it’s essentially the Internet archive. Since 2001, the Wayback Machine has viewed and stored any web page publicly available on the World Wide Web. Why am I highlighting this product? When RadicL Thoughts turned into Just Thoughts, I made the radical move of deleting all the previous writing I had published online since I started in 2018. The endeavor is explained in more detail in Just Thoughts #1 - Rebranding to “Just Thoughts”.
I also deleted all versions of my work stored on personal drives. However, The Wayback Machine is just one example of how anything publically posted online can never really be deleted and is most certainly exploited by someone in the era of AI.
The nice part is that you can still find content you may have lost in time. I probably have spent a year writing the content that went into creating one of my last posts on nicolasdolenc.medium.com - “Finding Balance and Purpose II”.
Furthermore, my writing now exists in inboxes, paper books, and hard drives stored by only a few, with less than a handful of people who have my thoughts written on paper, if you don’t count the birthday cards I’ve given or other handwritten poems… Now that I’ve talked enough about myself, what about you? Why are you doing what you do?
As humans, we’re always trying to cheat death until we accept our mortality, and even then, we still aspire to leave something behind for others to remember us. Every religion in the world is built around this fear and desire. Others would argue we’re biologically wired to have children as a need to replicate ourselves. In most cultures, we transfer our family names to outlive our existence.
We don’t walk around actively thinking about these thoughts but reflecting on them can be an exciting exercise when considering why people create products and companies.
What we created with a bit of Make & Buffer Magic
In the article linked above, "Just Thoughts #1 - Rebranding to Just Thoughts my cousin was also mentioned as someone looking for work in Finland. He ultimately got hired by the local bank Nordea, where he started a while ago, but before that, he helped me set up some automation for the Just Thoughts publication.
The reason for the setup
As I spent a lot of time writing long-format content, I wanted to use AI to help generate social content and drive engagement from other platforms to the Just Thoughts publication. I wasn’t using anything, but LinkedIn was on Substack's side for about three years, having deleted every social profile like Twitter, Instagram, and Facebook. I recommend that type of social media detox for everyone.
Nonetheless, after this writing exercise, I was ready to create the profiles but wasn’t prepared to start spending too much time on the platforms themselves. Hence, I thought if I spent 20-30 hours writing long-format content each week, AI could at least capture and post the content summaries for me.
The solution
If my “superpower” is writing, my cousin will geek out on data engineering and automation. Hence, I didn’t try to figure this out; he could whip something up much faster than I could. Here’s a recording of him explaining how it works.
In case your Portugues is rusty (we blend Portugues and English when we talk),
The “control center” is a Google spreadsheet. As seen in the video above.
You post a link to the published (Just Thoughts) post in one column. This would work with any webpage.
You can toggle each row between “ready,” “pending,” and “canceled.”
A row with a populated link and “ready” will push the webpage to a GPT on ChatGPT, which is configured to create tweets, threads, and LinkedIn posts for a particular audience (we’re targeting entrepreneurs who lead teams of 5-25 people who have kids). You can configure your GPT to cater to any audience on any platform while defining optimal character length.
The output from the GPT is funneled back to the spreadsheet (“the control center”). Each platform has its tab, and each GPT output has its row. Additional columns for “control of length” ensure the posts don’t exceed character limits if edited before toggling “ready” in the control column. We also added a “date posted” for future reference if we want to circulate content in the future, for example.
When a post was edited for publishing, you pressed “ready,” and it would post in order from top to bottom at predefined times. We chose noon, 4 pm, and 8 pm on Threads and X, while only once daily on LinkedIn.
The automation platform is called “make.com,” we used Buffer to control access to social media platforms and the timing of posts. Make is priced based on the number of operations, and buffer is priced based on the number of accounts and seats, with three accounts and one free seat. Additionally, you’re paying something to use chatgpt’s API, but we’re talking cents in thousands of operations.
Other automation platforms like IFTTT and Zapier exist, but Make’s power lies in its drag-and-drop UX/UI and intuitive no-code solutions for many use cases. Buffer likely has many rivals, but it was the first we ran across. We didn’t play around enough with the actual tool to have an opinion one way or the other. It served its purpose, worked, and was free for our use case of three outlets and one seat.
Learnings
At the end of the video, Luciano explained that we could train the GPT to produce better content, especially for LinkedIn, where I was already active. Hence, in the first two weeks of running the automation, I rewrote each GPT output to sound “more like me,” or rather, I didn't publish anything that didn’t sound like something I would write.
We created a separate tab where we put the original GPT output and the edited version. After two weeks, we asked the GPT to learn from the examples and added additional configurations.
The stats of new engagement from threads show that Twitter wasn’t yielding anything new or driving more views to the content. The LinkedIn posts had little engagement, much less than the ones I wrote. We weren’t using any other tricks on X and Threads you should be doing, like commenting, liking, and following profiles with similar audiences.
I wouldn’t say I like spending too much time online engaging on platforms other than LinkedIn and Substack. I habitually stopped using the control center and defaulted to posting things manually using Substack’s share features and manual writing. I just figured if I don’t have the resources to put in the (“human”) time, why bother? Especially when it’s not generating views, nor did it seem to add subscriptions. Albeit, we may be applying the wrong growth strategy without a loop.
However, as threads eventually created some views and X didn’t, I didn’t feel inclined to post anything on X, but I continued with some manual posting on threads. I could start posting something on Warpcast (an “on-chain” X rival), which seems significant in the poetry sector, and try Bluesky, even if it simply appears to be “the old Twitter” before Elon took over.
What do you think? Where should Just Thoughts be posting?
Need help in creating something similar?
However, as with other projects where he uses this same automation setup, he'll be happy to chat with you if you’re interested in exploring what can be automated or if you already have a use case but don’t know how to create or maintain it. He can help! Reach out to him on LinkedIn. I recommend checking out Taina Pereniemi’s consulting services for more extensive automation projects. I interviewed Taina previously on Just Thoughts. 
That’s all for this edition of Just Products. The next edition will feature the product landscape in the people domain, the tools the “people team” uses to run and scale their operations, and some related product opportunities.
As promised to the subscriber audience, here’s a take on what the original “hold my beer” moment could’ve looked like.
Just Products - Welcome to the new era of product creativity
What did you think? Do we need to consider what voices we use to listen to the content? Should we prefer the original creator over the AI-generated voice, and what’s your favorite “hold my beer” moment?
Who should see or hear about these ideas? Share if you’re inspired!
Just Thoughts is a reader-supported publication. To receive new posts and support our work, consider subscribing to gain exclusive content with each post.