I tried this with a non-fiction book I read recently and it honestly feels like a pretty generic LLM output. The summary roughly follows the text, but in many spots takes things out of context to a degree that the summary would be very misleading to anyone who hadn't already read the book.
The quotes at the start of each section were some of the worst offenders. I really have no idea what the first quote was referring to. It didn't relate to any of the themes of the book, or any of the following summary points, and I don't remember reading that line at all; it just seemed totally out of place.
I'm not really sure what I would use this service for. When I'm interested in a book that I haven't read it's generally the overarching concepts and takeaways that I want to know, not a spotty outline of the different sections of text. Perhaps students who want to skim through some assigned reading would find this useful. I'm also not that interested in Blinkist so I might not be the target audience here.
Hmm, which book was it? I can check. It is designed to complement the original work by allowing you to preview/review books before/after reading them, not to replace the need to read the original book.
As an ex-Blinkist I think the work here it's quite impressive.
From what I found based in some titles that I have read in the past and in comparison with Blinkist:
- The summaries here feels that has more "meat in the bone" in terms of bring complex ideas and have more bullet points
- Maybe it's about the prompting, but the language used you can definitely know that is generated by an LLM
- Having the direct quotes from the book after an initial definition of the theme of a chapter it's good
- The whole experience with the LLM kicking in on the fly, makes the whole experience quite bad; maybe you should consider have that pre-stored in some place
Honestly, I think your product it's 85% of theirs (and this is not something bad).
The pricing USD 79 for a lifetime it's OK, but without giving some kind of access or even make some library available it's quite hard to see someone paying for a library that you do not know the books that is in there.
From the legal perspective, I think you might be in some hot water since some publishers can dispute the fair use due to the fact that you're monetizing part of their content.
Okay, I really appreciate the feedback! Agree on the "meat in the bone," quotes, and LLM-like writing style (which is somewhat intended).
On-the-fly/cache: Yeah, I agree. We're only four months old—the number of on-the-fly summaries will go down significantly over time.
85%: What could make it 100%?
Pricing: I think that's fair. But considering Blinkist charges for $100+ for yearly, the $80 lifetime deal isn't that bad.
Legal: We actually drive a decent amount of traffic to Amazon, Audible, and Kindle. We're hoping to partner with the publishers and authors in the future whether in the form of revenue sharing or licsening.
Meaning you “generated” 73k “summaries” of which you are wholly unable to verify the content of because it would require you to have read the books and listened to/read the “summaries”.
What is with sloppybros and thinking quantity is more desirable than quality?
What motivated you to “remove” this fact from your website? Legitimate criticism?
> Meaning you “generated” 73k “summaries” of which you are wholly unable to verify the content of because it would require you to have read the books and listened to/read the “summaries”.
No, we generate summaries on the fly, too. Why would you generate all the summaries in advance when you can summarize real-time when a user clicks on it?
> What is with sloppybros and thinking quantity is more desirable than quality?
Have you read any of our summaries? Go ahead and read a summary of a book you've actually read and let me know what it's lacking. IMO, our summaries are way more thorough than human-written counterparts. And the structure is consistent across the entire catalog. In other words, we offer both quality and quantity.
> What motivated you to “remove” this fact from your website? Legitimate criticism?
No, just a matter of branding. We thought "AI" might add a cool factor, but everyone's AI nowadays, so we're just differentiating. Also, we didn't care to buy the ".ai" domain either. Most people already know it's AI-powered without us letting them know.
> Why would you generate all the summaries in advance when you can summarize real-time when a user clicks on it?
Well the most obvious reason is that the summary page would load faster. I clicked on a book summary and it took much longer to render than even a bloated web app usually takes.
73k summaries isn't that many. If your site gets any traction most of those summaries will be hit repeatedly and will have to generate anyway. Creating the summaries in advance would also allow you to test and verify the output, if you're interested in that.
Okay, that's fair. It isn't that many, but it's still 10x more than the industry leader, Blinkist, though. A good start.
As a small company, we currently need to sacrifice the first reader's experience for the sake of spreading out the costs over time. But every subsequent read is cached!
The most common first action people take on our site is reading a summary of a book they've already read to assess its quality themselves. I don't think they care whether it was written by OpenAI or a monkey, as long as it's good.
This is why no one buys books. Books are too long, too broad, the interface of books is the same since 868 CE and 90% of the text is just filler. Kindle brought some innovation into this space, but it's UI and support tools are still ridiculously cro magnon like.
The invisible hand that somehow forces authors to bloat their 3 paragraph gist into 200 pages kills the whole industry. I don't care if a book is only 15 pages long and costs $10, just leave the filler out or separate it with a blank page, then could come the 200 page long bs.
I used to buy books personal development, management and productivity books and a very sad transition in that space in the last 15 years is that most of the time the author needs to used a lot of anecdotes and make a narrative instead to do directly do the point.
I used to call that as a "Talebnization" of business/management writing that everyone uses more or less the same format to convey a idea:
- Some random quote to support the point
- 1 paragraph with the point
- 1 or 2 anecdotes or researched case about the point
- Some broad cherry-picked statistic to bring some rigor
- A success case due to the point
- Closing thoughts without any counterpoint, critique, or presence of any downside in the main point of the chapter.
I have mixed feelings. I generally agree—the industry tends to incentivize writers to fill a certain amount of space, much like college essays. However, there are quite a few books I've enjoyed in their entirety, discovering new insights when re-reading them.
Yeah, but I'm talking in a broad way, and would want the average people to read more books. Relevant books, something that clicks with them, but in order to do so, this current, outdated and sloggy format books have must go.
oh man, LLM summaries usually blow, particularly the longer the text they are summarising. have you heard the HN generated podcast? still better to read the actually post and comments.
It’s the very definition of autogenerated sometimes correct sometimes full of shit with no way to tell spam that isn’t even being honest with users that it’s AI nonsense.
I'm sorry to break this news to you, but your attitude ruins the internet/world more. Can you point out one instance of 'full of shit' content on the site?
I tried this with a non-fiction book I read recently and it honestly feels like a pretty generic LLM output. The summary roughly follows the text, but in many spots takes things out of context to a degree that the summary would be very misleading to anyone who hadn't already read the book.
The quotes at the start of each section were some of the worst offenders. I really have no idea what the first quote was referring to. It didn't relate to any of the themes of the book, or any of the following summary points, and I don't remember reading that line at all; it just seemed totally out of place.
I'm not really sure what I would use this service for. When I'm interested in a book that I haven't read it's generally the overarching concepts and takeaways that I want to know, not a spotty outline of the different sections of text. Perhaps students who want to skim through some assigned reading would find this useful. I'm also not that interested in Blinkist so I might not be the target audience here.
Hmm, which book was it? I can check. It is designed to complement the original work by allowing you to preview/review books before/after reading them, not to replace the need to read the original book.
I expected the books to be public domain, but no, it's just random (?) books from many different publishers. I hope you have a good legal team...
I believe the content falls under Fair Use. More details: https://docs.sobrief.com/en/articles/9905825-is-sobrief-infr...
Laundering a pirated book collection through OpenAI is a hell of a business model. Good luck!
Thank you for your kind feedback!
isn't that basically Blinkist
Did you read the title or site? It says they're summaries which are definitely fair use.
Now whether they're interesting or not is another question.....
As an ex-Blinkist I think the work here it's quite impressive.
From what I found based in some titles that I have read in the past and in comparison with Blinkist:
- The summaries here feels that has more "meat in the bone" in terms of bring complex ideas and have more bullet points
- Maybe it's about the prompting, but the language used you can definitely know that is generated by an LLM
- Having the direct quotes from the book after an initial definition of the theme of a chapter it's good
- The whole experience with the LLM kicking in on the fly, makes the whole experience quite bad; maybe you should consider have that pre-stored in some place
Honestly, I think your product it's 85% of theirs (and this is not something bad).
The pricing USD 79 for a lifetime it's OK, but without giving some kind of access or even make some library available it's quite hard to see someone paying for a library that you do not know the books that is in there.
From the legal perspective, I think you might be in some hot water since some publishers can dispute the fair use due to the fact that you're monetizing part of their content.
Okay, I really appreciate the feedback! Agree on the "meat in the bone," quotes, and LLM-like writing style (which is somewhat intended).
On-the-fly/cache: Yeah, I agree. We're only four months old—the number of on-the-fly summaries will go down significantly over time.
85%: What could make it 100%?
Pricing: I think that's fair. But considering Blinkist charges for $100+ for yearly, the $80 lifetime deal isn't that bad.
Legal: We actually drive a decent amount of traffic to Amazon, Audible, and Kindle. We're hoping to partner with the publishers and authors in the future whether in the form of revenue sharing or licsening.
> Please wait for the current search to finish.
This message modal appeared when I tried to select a book. I'm sure you know that's not the user experience anyone wants.
Good point! Easy fix.
OP here—happy to answer any questions! We’ve been working really hard since writing the first line of code 4 months ago.
As others already clarified that it is using an LLM do you allow users to report issues with summaries? I would think this might be important.
Yeah, great point. Will add that!
4 months, 73k summaries, 0 mentions of AI on the website/in the FAQ.
How are you doing it?
we used to mention AI in the title, but we removed it. we did list it on a bunch of ai directories like future pedia which does say we use ai
EDIT: just added a help article saying that we use ai
So, AI.
Meaning you “generated” 73k “summaries” of which you are wholly unable to verify the content of because it would require you to have read the books and listened to/read the “summaries”.
What is with sloppybros and thinking quantity is more desirable than quality?
What motivated you to “remove” this fact from your website? Legitimate criticism?
> Meaning you “generated” 73k “summaries” of which you are wholly unable to verify the content of because it would require you to have read the books and listened to/read the “summaries”.
No, we generate summaries on the fly, too. Why would you generate all the summaries in advance when you can summarize real-time when a user clicks on it?
> What is with sloppybros and thinking quantity is more desirable than quality?
Have you read any of our summaries? Go ahead and read a summary of a book you've actually read and let me know what it's lacking. IMO, our summaries are way more thorough than human-written counterparts. And the structure is consistent across the entire catalog. In other words, we offer both quality and quantity.
> What motivated you to “remove” this fact from your website? Legitimate criticism?
No, just a matter of branding. We thought "AI" might add a cool factor, but everyone's AI nowadays, so we're just differentiating. Also, we didn't care to buy the ".ai" domain either. Most people already know it's AI-powered without us letting them know.
> Why would you generate all the summaries in advance when you can summarize real-time when a user clicks on it?
Well the most obvious reason is that the summary page would load faster. I clicked on a book summary and it took much longer to render than even a bloated web app usually takes.
73k summaries isn't that many. If your site gets any traction most of those summaries will be hit repeatedly and will have to generate anyway. Creating the summaries in advance would also allow you to test and verify the output, if you're interested in that.
Okay, that's fair. It isn't that many, but it's still 10x more than the industry leader, Blinkist, though. A good start.
As a small company, we currently need to sacrifice the first reader's experience for the sake of spreading out the costs over time. But every subsequent read is cached!
> No, we generate summaries on the fly, too.
Yikes, so the summaries change with each viewing?
Information that changes each time you read it is by definition “unreliable”.
> Go ahead and read a summary of a book you've actually read and let me know what it's lacking.
Work for a sloppyjoe? And for free?!
Come on. I think you have clearly grown a little too accustom to exploiting other peoples’ work.
Hey, just have your AIs check them all. Right?
It's 2024, it's not worth the time it takes to stop and wonder if it's LLM shovelware. It always is.
true
Which LLM is used for the book summaries and which TTS for the audio?
llm: prefer not to answer tts: open ai
How do you expect anyone to trust the summaries if you are not even going to disclose which AI model generated it?
The most common first action people take on our site is reading a summary of a book they've already read to assess its quality themselves. I don't think they care whether it was written by OpenAI or a monkey, as long as it's good.
Hello and great site! Just a note that the "business" isn't spelt correctly. I'm enjoying it ATM. Tks.
Oh, lol. Just fixed it!
Do you purchase the book before making a summary? How does that work?
Yes, we try to borrow as often as possible and buy when necessary
This is why no one buys books. Books are too long, too broad, the interface of books is the same since 868 CE and 90% of the text is just filler. Kindle brought some innovation into this space, but it's UI and support tools are still ridiculously cro magnon like.
The invisible hand that somehow forces authors to bloat their 3 paragraph gist into 200 pages kills the whole industry. I don't care if a book is only 15 pages long and costs $10, just leave the filler out or separate it with a blank page, then could come the 200 page long bs.
I used to buy books personal development, management and productivity books and a very sad transition in that space in the last 15 years is that most of the time the author needs to used a lot of anecdotes and make a narrative instead to do directly do the point.
I used to call that as a "Talebnization" of business/management writing that everyone uses more or less the same format to convey a idea: - Some random quote to support the point
- 1 paragraph with the point
- 1 or 2 anecdotes or researched case about the point
- Some broad cherry-picked statistic to bring some rigor
- A success case due to the point
- Closing thoughts without any counterpoint, critique, or presence of any downside in the main point of the chapter.
Do that for 20 chapters and you have a book.
I have mixed feelings. I generally agree—the industry tends to incentivize writers to fill a certain amount of space, much like college essays. However, there are quite a few books I've enjoyed in their entirety, discovering new insights when re-reading them.
Yeah, but I'm talking in a broad way, and would want the average people to read more books. Relevant books, something that clicks with them, but in order to do so, this current, outdated and sloggy format books have must go.
This is clean, nice work!
Thank you!
oh man, LLM summaries usually blow, particularly the longer the text they are summarising. have you heard the HN generated podcast? still better to read the actually post and comments.
are you referring to this one? https://news.ycombinator.com/item?id=35831177
Thanks, I hate it and this is exactly the kind of thing I fear is ruining the internet.
How would this ruin the internet?
It’s the very definition of autogenerated sometimes correct sometimes full of shit with no way to tell spam that isn’t even being honest with users that it’s AI nonsense.
I'm sorry to break this news to you, but your attitude ruins the internet/world more. Can you point out one instance of 'full of shit' content on the site?
this doesn't prevent you from reading the books though?