Hacker News Reader

Hacker News

Comment by FrustratedMonky | Hacker News Reader

Full thread

FrustratedMonky

I clicked on Gladiator.

Is this really all AI???? This whole Gladiator Podcast?

I can't tell, it is totally believable.

Really, this is too good, can this be confirmed if it is AI or not?

View on HN

Definitely too good! The only way I can tell they're AIs is they don't pause or hesitate in their banter. Their ability to talk without thinking is uncanny.

And they don't interrupt each other

Look up google notebooklm. 100% ai. It's great, ive used it to learn some complex topics I couldn't get my head around.

But after listening to a good 10 hours of it, I'm seeing the equivalent of ai slop and repetitive patterns in all of them :(

Yes its notebooklm.

So all text and voices? An AI is doing both sides of conversation, just off the cuff from reading the script?

This is amazing.

yea. everything. try it out (point it at a website or something)

Yeah I have a hard time believing this. Even OpenAI's voices don't sound this natural with the pauses in the speech.

Maybe because the OpenAI App where you talk to it is responding in Realtime.

But the NotebookLM is like a recorded pod-cast, so it can think about it before laying down a track?

It's because oai are going for low latency. I've got pause free working locally by running the tts model after the llm has finished generating a huge chunk of the response.

And fine tuning on 15 minutes of the voice you're targeting makes it sound natural

Ok, so on further thought I think I can tell that they're AI now. But it's subtle. The pace of the conversation is just a bit too quick for natural human speech. They're talking about history but the female voice especially, is very quick to respond with points as soon as the male voice is finished. It's clear that the AI doesn't factor in the time it takes for someone to think momentarily and formulate a thought.

Funny that it is too smooth, quick, and adding a pause might be more realistic.

But also, some people talk fast.

They'd have to add some random pausing.

It's not so much the speed of the speech itself. The speech sounds natural and there are natural sounding inflections. But the gap between the speakers is almost nonexistent. It's quite exhausting to listen to because your own brain is trying to think about the content but there's no time to do it in parallel with the hosts, since they're not actually thinking.

I'm not seeing any voice options on NotebookLM.

Did notebooklm produce the text, and then they had real people read it out loud?

You upload a document then click the Generate Podcast button.

Reply on news.ycombinator.com