Live data from Hacker News

Google Illuminate: Books and papers turned into audio

illuminate.google.com

241–250 of 251 posts

Re: Google Illuminate: Books and papers turned into audio

#241

Earlier quoted context omitted.

I listened to a lot of current AI „podcasting“ tools and wh ok me the voice is 95% perfect it does have its issues: - suddenly speeding up or slowing down - mispronunciation of non-standard words - weird pauses

Having listened to a great many podcasts and interviews ... these are all very much problems with human-embodied voices as well. (The number of SV types who talk as if they're on coke / meth / speed is ... nuts. A certain A-Z lead character comes to mind. Piketty is another. It'd be less problematic if they weren't constantly tripping over their own words, but they are.)

Example of a human narration with a number of distracting issues:

https://hakaimagazine.com/features/the-big-baltic-bomb-clean...>

Throughout the piece, the narrator reads "ordnance" as "ordnances" (the word is both singular and plural), some awkward emphasis within phrases, and odd pronunciations ("malevolent" stands out). On the other hand, her pronunciation and accent of German words and place-names is excellent.

Re: Google Illuminate: Books and papers turned into audio

#242
post #240

Earlier quoted context omitted.

I'm building this at https://reasonote.com/app/login Sign up and I'll let you in very soon.

Can i try it as well? I have a college professor buddy of mine as well who loves this kinda stuff and am sure would be happy to be an early tester

Absolutely! Will be letting lots of folks in tomorrow.

Re: Google Illuminate: Books and papers turned into audio

#243

This is really cool, and it got me thinking - is there any missing piece to creating a full AI lecturer based on this? What I'm thinking of is that I'd input a pdf, and the AI will do a bit of preprocessing leading to the creation of learning outcomes, talking points, visual aids and comprehension questions for me; and then once it's ready, will begin to lecture to me about the topic, allowing me to interrupt it at a…

https://www.rebind.ai/

Re: Google Illuminate: Books and papers turned into audio

#244

Earlier quoted context omitted.

Would you listen to an auto-generated podcast? Seems like removing the humans from the equation kind of defeats the purpose.

A lot of our customers use us [0] for that, it works pretty well if executed properly. The voiceovers work best as inserts into an existing podcast. If you see the articles of major news orgs like NYT, they often have a (usually) machine narrated voiceover. [0] https://narrationbox.com

Looks good! Do you guys have an API?

Re: Google Illuminate: Books and papers turned into audio

#245
post #194

By now, we can find thousands of hours of discussions online about popular papers such as "Attention is All You Need". It should be possible to generate something similar without using the paper as a source -- and I suspect that's what the AI does. In other words: I suspect that the output is heavily derivative from online discussions, and not based on the papers. Of course, the real proof would be to see the output…

There are much newer papers shown than "Attention is All You Need" (all of them?) and much less talked about (probably all of them, too). It shouldn't be surprising that a LLM is able to understand a paper, just upload one to Claude 3.5 Sonnet.

Of course, "Attention is All You Need" was one of the most discussed papers in our field; there are entire podcast episodes dedicated to it, so it should be easy for a LLM to create a new one.

For all the other papers, assuming they were impactful, they must have been referred by others, highlighting what their contribution is, what is controversial, etc.

In other words: the LLM doesn't have to "understand" the paper; it can simply parrot what others have been saying/writing about it.

(For example: a podcast about Google Illuminate could use our brief exchange to discuss the possible merits of this technology.)

Re: Google Illuminate: Books and papers turned into audio

#246
post #239

Earlier quoted context omitted.

First off, why have you moved the goal-posts, expecting LLMs to be not just at human level, but at subject matter expert level? And second, I would appreciate recommendations of good debates where both sides have a lot to offer and don't fall into errors; we do need more of those.

Because that's basic assumption. I'd want a discussion about academic papers from subject matter experts otherwise reading it aloud alone is not of any value let alone commentary on it with no understanding of it whatsoever.

There's a spectrum of value here. I don't think we're at the SME level yet, but I'd say we are at least at the level of presentation you'd get at an undergrad seminar module - and I already see significant value here.

Re: Google Illuminate: Books and papers turned into audio

#247
post #239

Earlier quoted context omitted.

Because that's basic assumption. I'd want a discussion about academic papers from subject matter experts otherwise reading it aloud alone is not of any value let alone commentary on it with no understanding of it whatsoever.

There's a spectrum of value here. I don't think we're at the SME level yet, but I'd say we are at least at the level of presentation you'd get at an undergrad seminar module - and I already see significant value here.

True. It is not totally useless. There's a lot of utility.

Re: Google Illuminate: Books and papers turned into audio

#248
post #39

Earlier quoted context omitted.

Maybe not a podcast, but I've often wished I could listen to a paper or an article while on a long drive

Could be me, but the amount of attention I need to reserve in order to properly read and understand a technical paper makes this idea rather scary.

Yeah technical papers probably wouldn't be a good use case. And definitely not what I had in mind. I was thinking about thinkpieces and older books that missed the audiobook trend

Re: Google Illuminate: Books and papers turned into audio

#250

Earlier quoted context omitted.

A lot of our customers use us [0] for that, it works pretty well if executed properly. The voiceovers work best as inserts into an existing podcast. If you see the articles of major news orgs like NYT, they often have a (usually) machine narrated voiceover. [0] https://narrationbox.com

Looks good! Do you guys have an API?

Yes, coming soon
Post reply on HN