Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Google Illuminate: Books and papers turned into audio

illuminate.google.com

Post reply on HN
Print view View original

81–90 of 251 posts

Re: Google Illuminate: Books and papers turned into audio

#81

Post by jhickok »

paxys wrote: post #6
fny wrote: post #2

Very clever use case. I'm presuming the set up here is as follows: - LLM-driven back and forth with the paper as context - Text-to-speech Pricing for high quality text to speech with Google's studio voices run at USD 160.00/1M count. And given the average 10 minute recording at the average 130 WPM is 1,300 words and at 5 characters per word is 6500, we can estimate an audio cost of $1. LLM cost is probably about the…

Retail pricing != Google's actual cost.

I would actually be surprised if companies are focusing on profit at this stage.

Re: Google Illuminate: Books and papers turned into audio

#83

Post by swyx »

fallinditch wrote: post #43
freefaler wrote: post #3

Great idea. I wonder how long until we'd see a lot of "autogenerated" podcasts with syndicated advertising inside spamming the podcast space. Like with robovoiced videos on YT reading some scraped content.

Wondercraft have been offering this service for a while, and produce some of their own auto-generated podcasts including the Hacker News Recap which does an excellent job of summarizing the most engaged posts on HN. https://www.wondercraft.ai/our-podcasts

also for papers there is https://papersread.ai/ which does not get nearly enough attention imo (the reading is meh, but the curation is ace)

Re: Google Illuminate: Books and papers turned into audio

#84

Post by banach »

I can see this working reasonably for text that you can understand without referring to figures, and for texts for which there is external content available that such a conversation could be based on. For a new, say, math paper, without prose interspersed, I’d be surprised if the generated conversation will be worth much. On the other hand, that is a corner case and, personally, I suspect I will be using this for the many texts where all I need is a presentation of the material that is easy to listen to.

Re: Google Illuminate: Books and papers turned into audio

#85

Post by aanet »

What a fantastic idea! Great way to learn about those pesky research papers I keep downloading (but never get to reading them). I tried a few, e.g. Attention is All You Need, etc. The summary was fantastic, and the discussion was, well, informative.

Does anyone know how the summary was generated? (text summarization, I suppose?) Is there a bias towards "podcast-style discussion"? Not that I'm complaining about it - just that I found it helpful.

Re: Google Illuminate: Books and papers turned into audio

#86

Post by aanet »

vanishingbee wrote: post #71
dlisboa wrote: post #12

One problem I see with this is legitimizing LLM-extracted content as canon. The realistic human speech masks the fact that the LLM might be hallucinating or highlighting the wrong parts of a book/paper as important.

Happens in the very first example: [Attention is All You Need - 1:07] > Voice A: How did the "Attention is All You Need" paper address this sequential processing bottleneck of RNNs? > Voice B: So, instead of going step-by-step like RNNs, they introduced a model called the Transformer - hence the title. What title? The paper is entitled "Attention is All You Need". People are fooling themselves. These are stochastic p…

I had the same exact thought - "Did this summary mis-represent the title??" Indeed, it did. However, I thought the end2end implementation was decent.

> These are stochastic parrots cosplaying as academics.

LOL

Re: Google Illuminate: Books and papers turned into audio

#88

Post by swyx »

wintermutestwin wrote: post #65
bitshiftfaced wrote: post #58

Occasionally there's a podcast or video I'd like to listen to, but one of the voices is either difficult to understand, or in some way awful to listen to, or maybe the sound quality is really bad. It would be nice to have a an option for an automatically redubbed audio.

I sure do wish podcasters would learn about compression. I am constantly getting my ears blown out in the car from a podcast with multiple speakers who are at different volumes.

podcaster here. what does compression have to do with it? youre just talking about different levels from diff mics

Re: Google Illuminate: Books and papers turned into audio

#89

Post by leobg »

I made something like this for my kids:

1. Take a science book. I used one Einstein loved as a kid, in German. But I can also use Asimov in English. Or anything else. We’ll handle language and outdated information on the LLM level.

2. Extract the core ideas and narrative with an LLM and rewrite it into a conversation, say, between a curious 7 year old girl and her dad. We can take into account what my kids are interested in, what they already know, facts from their own life, comparisons with their surroundings etc. to make it more engaging.

3. Turn it into audio using Text-to-Speech (multiple voices).

Re: Google Illuminate: Books and papers turned into audio

#90

Post by flakiness »

leobg wrote: post #89

I made something like this for my kids: 1. Take a science book. I used one Einstein loved as a kid, in German. But I can also use Asimov in English. Or anything else. We’ll handle language and outdated information on the LLM level. 2. Extract the core ideas and narrative with an LLM and rewrite it into a conversation, say, between a curious 7 year old girl and her dad. We can take into account what my kids are intere…

How do you get the source data (text) from a book? To me it is the major roadblock for LLM-based commercial content consumption.
Post reply on HN