Mind Readings: Same is Lame Even in Generative AI

In today's episode, Jay Bear, my friend Jay Bear, famously said, same is lame. And there's this is an incredibly true thing. When everyone's doing the same thing, when everything sounds the same, looks the same, feels the same, it's lame. There's nothing setting you apart. There's no brand differentiation.

There's nothing that makes you go, wow, that's really different and cool. When you create the same thing everyone else creates, there's there's no literally nothing that sets you apart. People look across the landscape and see a sea of sameness, and they don't know who to buy from because they can't tell. The latest AI craze is marketers mostly using a very good research tool, an excellent product called Notebook LM from Google. Notebook LM is a rag locked gender AI tool.

And what that means is that this tool will only answer questions of data that you give it. You upload some documents and then you ask questions of it. So if you're using, if you're doing something, say like academic research, where you absolutely positively cannot afford to have the machine making things up, hallucinating, the technical term. This is a great tool to use. It is a fantastic tool to use because it will you can ask it questions and it will say, um, no, I can't answer the question.

So I could let me ask, you know, uh, what are some ways that this paper addresses long COVID? And NoBook LM will look at the paper that I uploaded, which has nothing to do with this. It's all about model tuning and stuff. And it will say, uh, this is this contains nothing about long COVID. Sorry.

Right. This is what you want. This is a great tool. This is what you want when you're doing high-risk research because it will say absolutely nothing that's not there. But what people have glommed on to is the audio overview.

And what the audio overview does is it allows you to listen to a generated podcast like two-host conversation of the machine summarizing your paper. Let me try and get a copy of this into here so we can listen to it. Snippet add some music, which is not music at all. Okay. I'm gonna put on my headphones too, so that I can hear as the our synthetic podcast begins.

It's kind of wild when you think about it, right? Like we've got these AIs, these large language models spitting out text, code, you name it, but uh it's like that final polish is missing, you know, like that step back and saying, hmm, okay, this could be a little better. Yeah, and it's funny, that's something we humans just kind of we take it for granted. We don't always nail it the first time around, but we can usually spot when something's off in our own work, yeah. And that's like the holy grail right now with AI intrinsic.

That's pretty good, right? That is pretty good. That is so behind the scenes, this is speculation because I don't know this for sure, but uh behind the scenes, what's likely happening is that Google is taking the document itself, passing it through a prompt in Gemini 1.5, which is the same engine that powers the rest of Notebook LM, and then generating uh those two voices in a conversation. So it builds a script, and then the two voices read it out. Those two voices are these journey voices when from Google's text to speech product.

They're very recognizable. If you know the text to speech uh market at all, and you know the different providers, uh here. Let me just show here. So this is Google Text to Speech. Here is the three of the journey voices that, yeah, when you listen to the demos, like, oh, I recognize the I recognize those voices.

Yep, you sure do, because Notebook LM uses that as the way to bring to life those uh summaries. So what's with the problem with this? It seems like it's a great application technology. It is, it's a phenomenal application of the technology. What is happening is that a lot of very lazy marketers um are saying, hey, I can just put a document in this thing and it'll generate a podcast for me.

Now I'm a podcaster. Which, yeah, that's fair, but it's all the same. There are hundreds of these deep dive expert podcasts that have popped up in the last two weeks on the podcasting service of your choice that are all this. It's all notebook LM readouts. They're all the same, they all sound the same.

And all these you know, marketing AI experts encouraging everyone to use the exact same utility to create content makes for exactly the same sound and content all the time. And so marketers have created this sea of sameness. When every informational podcast sound is the same, two neural voices from from Google TTS. There's nothing in this that's a part of your brand at all. It is just the same voices.

Now, you could replicate this. And I would encourage you if you are technically inclined to actually try replicating this. I did last week just for fun. You take a language model that summarizes a document, you have it write a script, you have uh the TTS model of your choice that has supports multiple voices, read the document aloud, and then you save that to an audio file. Uh, in fact, earlier this week I built one that was of a virtual focus group where it's me, the human, and AI voices.

There are already open source projects that replicate notebook LM's functionality but give you a lot more control. Like how long do you want the summary to be? You don't get a choice. What voices do you want to use? Uh, if you want to use a tool like 11 Labs and use your own voice, which I would encourage that these are ways to do this.

Again, there is a market for personality-free audio content, right? No one is expecting a powerful personal brand uh reading the weather report, right? That is a case where you absolutely positive can just have it deliver in a in a purely machine generated voice the content because the content there is useful and valuable. But if you are trying to build a brand, if you're trying to build differentiation, if you're trying to build something that is leveraging your unique voice and your unique points of view and your unique prompts, this is not the tool to use. This is not the tool you use.

The good news is this it is easier than ever to stand out today when everyone else is pushing the easy button and making an ocean of same-sounding content. That's gonna do it for this episode. Thanks for tuning in. Talk to you soon. If you enjoyed this video, please hit the like button.

Subscribe to my channel if you haven't already, and if you want to know when new videos are available, hit the bell button to be notified as soon as new content is live.