diminikolaou

A year after it's last episode, we're bringing back Hacker News Recap. We launched a new feature* last week that made the automation of this very easy so we had no reason of holding it back anymore.

Hope you guys like it.

*In case you are interested about the feature, this is it. https://www.wondercraft.ai/wonda

This is cool. I can see the anti-monopoly of OpenAI argument, but apart from that is there a strong argument of being multi-LLM for a Codex-like agent?

Thank you for this write up - very insightful.

Why not just use GPT-4?

Can't find a way to connect. Ping at hn@wondercraft.ai - would be keen to get your thing going.

Thank you!

You can definitely use the platform for audiobooks. It's not native to importive an epub yet, but you can use the UI or the API to audify your book.

Awesome to hear!

Podcasts and Youtube Shorts have very different view count, I'm sure you know that.

In any case, it's clear that this podcast is not for you and that's ok.

Appreciate you taking the time to give us thoughtful feedback.

There is no question that authors delivering the work themselves is preferable. But that option is rarely available (only 20% of NYT Best Sellers have audio books read by the authors).

| "you’ll just end up moving this information from being locked in prose to being locked in sound"

I don't think that everyone shares this opinion. Your fellow YC Batchmates are very happy that Paul Graham's essays are now in audio format as an example. More than 1000 founders have subscribed to the feed.

Thank you for taking the time to go through this. I agree that there is room for improvement and your comment makes it clear what we can improve. But I disagree with the sentiment that it does not add value. There are 1k people that tune in daily to listen to this that would agree with me: https://op3.dev/show/f77aea62-97e5-5cce-92c6-9464e51c30c6

Definitely will add transcript. The HN Recap has been a side thing for us while we have been building the Podcast Creation platform, but we will do better.

Dming you on twitter, to find “The Thing”

Yeah that’s write, we highlight that the content should come from human creativity.

Wrote some stuff here about comparison to Descript:

https://news.ycombinator.com/reply?id=37089928&goto=item%3Fi...

For substackers that don’t have a podcast, only have a written format, this tool has given them great leverage as they can easily provide more value to their audience by giving them an audio version. There’s over 100 of those on the platform currently.

For substackers that already have a podcast, that’s a separate question. Ultimately it depends on if they think that they can use the time the put in for the recording and editing process to some other creative process and use Wondercraft for generating the audio. Haven’t seen that many existing Substack podcasts for this use case to become that prevalent.

11 is not our competitor, they are our partner and model provider. We’ve built all of the podcast related infrastructure (publishing, show notes, video etc) on top of their text to speech capabilities.

You’re not the only one

We highlight that the creativity comes from the human. The text to speech is the AI Component. There is a script assistant that can help you adjust the tone of the text, but the creativity always comes from human.

Thank you!

We’re adding all of the essays, should be all done by end of August.

I actually think I disagree with your last sentence.

It's not that the creator can't be bothered, but that the tool unlocks features that are just not feasible for a creator. Think of Hacker News Recap for example - a 20' episode has been generated every single day for the past 122 days. That's extremely hard for a person to do. So the consistency that this enables is one feature.

Automatic translation and video generation are other features that using this method of podcast creation lead to. So it's not necessarily that the "creator can't be bothered to"... there's inherently more value in some use cases to create a podcast this way over the traditional method.

It can be any voice the user chooses, but by default it's the same voice as the original language - all voices speak all languages.

PG has not approved for this to be done with his own voice, which is why we're using a custom voice (Anna).

Sorry what is exactly your claim here? That it is legal to use and monetise off of someone else's voice? Using Morgan Freeman's or Joe Rogan's voice for example to host your podcasts?

Strong words, but I understand where you’re coming from. I get your point on people liking authentic stories from people and that’s the value of podcasts. There’s many different formats of podcasts and interviews/conversations are just 20% of overall podcasts. I agree with you that for the time being those formats are best done person to person. But for all other podcasts (e.g news rundowns, narrations etc), podcasts generated on Wondercraft on a daily basis are proving that as long as human creativity is there podcasts created using TTS and other AI tools can be very engaging and appealing for many.

There is so much written content currently confined in written format and we have built an app that makes it very easy to convert it to engaging audio. Paul Graham essays is a prime example.

If that necessarily means there is also some “fake” podcasts, that no one will listen to anyway, I think it’s a fair exchange.

So although I understand your position, no, we are by no means making the world worse.

Many educators use Wondercraft right now as it also generates video which helps students. Check out if can be helpful to you.

In relation to HN Recap, I agree that it struggles with terms, but I’d argue that the results show that the stories are captured well.

I get your point on people liking authentic stories from people and that’s the value of podcasts. There’s many different formats of podcasts and interviews/conversations are just 20% of overall podcasts. I agree with you that for the time being those formats are best done person to person. But for all other podcasts (e.g news rundowns, narrations etc), podcasts generated on Wondercraft on a daily basis are proving that as long as human creativity is there podcasts created using TTS and other AI tools can be very engaging and appealing for many. That is ultimately how the flooding of such content is regulated as well, people will only listen to content that has thought behind it so creators will not find much value in “auto generated” podcasts.

For the time being, realistically it’s cost prohibitive for users to generate slop without expectation for some return, so that’s a current natural mechanism.

As the prices decrease, due to model improvement, best/worst practice guides, and promotion of successful use cases may act as an additional mechanism to prevent such cases.

Realistically though, given that we’re lowering the bar of podcast creation to effectively the level of blog creation there might be an increase in slop content. I think that’s a necessary problem when you lower the barrier, but there will always be proper curation in the aggregator level. I think HM is the perfect example. Internet has made blogging so easy and there is so much slop. But it goes unnoticed given the curation.

Thank you!

This is copyright infringement, not ok to be using and monetising off of someone else’s voice. Over the next months there are plans for way tighter controls on this so I’d expect that the “using the voices of the original speakers” feature will not be available, unless a monetisation method is developed.

Yeah hard to automate, but that’s why we built the UI to give the controls to everyone to make it as elaborate and high quality they want.

Thanks for kind words!