1timspalding
We've released a new, natural-language work search, called "AI Search."
Try it out: https://www.librarything.com/search.php?searchtype=ai
Examples:
The idea here is to use AI (specifically GPT3) to surface books. But because GPT3 is prone to "hallucinations," we wash everything through what LibraryThing knows about books. It might still be wrong, but the books are real. The result is a fairly powerful way to find books you can't find by ordinary means, like title and author. (No, I don't think it replaces tags, subjects and so forth either, and it definitely doesn't replace recommendations, but it's good for some things.)
There are some other notes on the pop-up when you enter. Notably, it can't find most books published after September 2021.
As many know, I'm an "AI skeptic," at least among developers. ChatGPT is wild, but I don't see as many uses yet as some do. More importantly, I'm worried about its tendency to lie ("hallucinations") and the ways it will be misused by bad actors, for example to generate spam. As far as I'm concerned, however, this is a pretty wholesome use, and not without value.
What do you think?
Try it out: https://www.librarything.com/search.php?searchtype=ai
Examples:
- female decoders during wwii
- What's the Greek play where someone is murdered in the bath?
- novel about Jesuits in space
- classic novel about going down the Mississippi on a raft
- nonfiction book about why Europe conquered the world
- dinosaurs come to life on an island
The idea here is to use AI (specifically GPT3) to surface books. But because GPT3 is prone to "hallucinations," we wash everything through what LibraryThing knows about books. It might still be wrong, but the books are real. The result is a fairly powerful way to find books you can't find by ordinary means, like title and author. (No, I don't think it replaces tags, subjects and so forth either, and it definitely doesn't replace recommendations, but it's good for some things.)
There are some other notes on the pop-up when you enter. Notably, it can't find most books published after September 2021.
As many know, I'm an "AI skeptic," at least among developers. ChatGPT is wild, but I don't see as many uses yet as some do. More importantly, I'm worried about its tendency to lie ("hallucinations") and the ways it will be misused by bad actors, for example to generate spam. As far as I'm concerned, however, this is a pretty wholesome use, and not without value.
What do you think?
2MarthaJeanne
I suggest trying it on books that Name That Book hasn't found.
3timspalding
>2 MarthaJeanne:
We did a fair amount of that, actually. It gets some of the easier ones. But not most of them. Experiments welcome, though.
We did a fair amount of that, actually. It gets some of the easier ones. But not most of them. Experiments welcome, though.
4paradoxosalpha
Huh. "novel about Jesuits in space" should be A Case of Conscience, but the search doesn't deliver.
5timspalding
>3 timspalding:
I know! But it gets the Sparrow and its sequel. However, try https://www.librarything.com/search.php?search=more+novels+about+Jesuits+in+spac...
It's currently "stateless." That means each query is separate, with no memory of the last. And it tends to return only a few books. You can juice it for more by asking for "at least 10 books" or such.
I know! But it gets the Sparrow and its sequel. However, try https://www.librarything.com/search.php?search=more+novels+about+Jesuits+in+spac...
It's currently "stateless." That means each query is separate, with no memory of the last. And it tends to return only a few books. You can juice it for more by asking for "at least 10 books" or such.
6Aquila
My eyes dropped to the date in my browser and no, its not an April Fool's joke.
And it makes sense and will be useful.
And it makes sense and will be useful.
7norabelle414
I am also skeptical about the usefulness of natural-language AI but this seems fine.
8amanda4242
>4 paradoxosalpha: A Case of Conscience comes up if you search "novels about Jesuits in space."
9timspalding
>8 amanda4242:
Indeed, if we weren't caching them for a time, it might come up if you searched the one that failed again. Large language models are twitchy and don't always answer the same way twice. Mysteries of the universe…
Indeed, if we weren't caching them for a time, it might come up if you searched the one that failed again. Large language models are twitchy and don't always answer the same way twice. Mysteries of the universe…
11Keeline
Didn't we have a lively discussion where the same Tim placed a terms of service rule against the use of AI-generated (specifically ChatGPT) reviews?
I think some clarity is in order of why one use is appropriate and another is not.
James
I think some clarity is in order of why one use is appropriate and another is not.
James
12klarusu
This is a really interesting take on it. I’m really curious because we’re deep into the implications of LLMs and other generative AIs for education right now. One of the things that has been raised about GPT, apart from the obvious ‘hallucinations’ issue, is that there are inherent biases being replicated with LLMs - basically because the content it ‘learns’ from still carries strong biases towards information about non-marginalised groups, GPT reinforces this by making it objectively harder to get it to return results that aren’t excluding under-represented groups. Do you find that there’s any ‘bias’ in the books it finds vs those it doesn’t?
13reading_fox
Is this going to make all the treasure Hunts easier?!
14rodneyvc
>6 Aquila: Oddly, I did the same
>13 reading_fox: Possibly - I ran the clues for the 5 works in the 2023 Valentine hunt through the search, and it got 3 (Q2, Q11, Q14). I didn't bother with the ER novel, as it is too recent.
>13 reading_fox: Possibly - I ran the clues for the 5 works in the 2023 Valentine hunt through the search, and it got 3 (Q2, Q11, Q14). I didn't bother with the ER novel, as it is too recent.
15SandraArdnas
>11 Keeline: Seems obvious to me, one is merely a search of existing LT data, the other is supposed to be user generated content
16andyl
Hmm just tried "best british science fiction of the last 5 years" and got
The Book of Strange New Things by Michael Faber. Naturalised British. Book published 2014
A Long Way To A Small Angry Planet by Becky Chambers. American. Book published 2014.
Blackfish City by Sam J Miller. American. 2018
The Testaments by Margaret Atwood. Canadian. 2019
Adrift On the Sea of Rains by Ian Sales. Ian is British. Book published 2012.
So 2 out of 5 on nationality (neither of them books I would have expected.), 0 out of 5 when you add in the date criteria.
"best science fiction of the last 5 years by a british writer"
I get the Chambers and the Atwood again along with
The City We Became by N. K. Jemisin. American
The Memory Police by Yoko Ogawa. Japanese.
So 0 out of 4.
For those that are wondering
"best british science fiction published after 2017"
still brings back Atwood, Jemisin, Ogawa and Tamsyn Muir in its list of 6 works. The other 2 are books by Adrian Tchaikovsky and M. John Harrison - who really are British.
So 2 out of 6.
"best science fiction by a british writer published after 2017"
brings back Atwood, Jemisin, and adds in John Scalzi and Kameron Hurley with Adrian Tchaikovsky being the only British writer in the list.
So 1 out of 5.
At least all the books are science fiction. So it does show that one should take the answers with a large pinch of salt.
The Book of Strange New Things by Michael Faber. Naturalised British. Book published 2014
A Long Way To A Small Angry Planet by Becky Chambers. American. Book published 2014.
Blackfish City by Sam J Miller. American. 2018
The Testaments by Margaret Atwood. Canadian. 2019
Adrift On the Sea of Rains by Ian Sales. Ian is British. Book published 2012.
So 2 out of 5 on nationality (neither of them books I would have expected.), 0 out of 5 when you add in the date criteria.
"best science fiction of the last 5 years by a british writer"
I get the Chambers and the Atwood again along with
The City We Became by N. K. Jemisin. American
The Memory Police by Yoko Ogawa. Japanese.
So 0 out of 4.
For those that are wondering
"best british science fiction published after 2017"
still brings back Atwood, Jemisin, Ogawa and Tamsyn Muir in its list of 6 works. The other 2 are books by Adrian Tchaikovsky and M. John Harrison - who really are British.
So 2 out of 6.
"best science fiction by a british writer published after 2017"
brings back Atwood, Jemisin, and adds in John Scalzi and Kameron Hurley with Adrian Tchaikovsky being the only British writer in the list.
So 1 out of 5.
At least all the books are science fiction. So it does show that one should take the answers with a large pinch of salt.
17timspalding
>16 andyl: Definitely. It's similar when you ask it for recommendations per se. LibraryThing's recommendation gnomes aren't going to be put out of a job anytime soon.
Note, however, that for GPT3/4 the world ends in 2021, so "last five years" may mean something different to it.
It does awards okay, but not enough of them: https://www.librarything.com/search.php?search=Hugo+Award+2019&searchtype=ai...
Note, however, that for GPT3/4 the world ends in 2021, so "last five years" may mean something different to it.
It does awards okay, but not enough of them: https://www.librarything.com/search.php?search=Hugo+Award+2019&searchtype=ai...
18andyl
>17 timspalding:
Sure I knew GPT-3's world ends in 2021 - but still five years before that was after 2014. I would have expected better results on the British part of the query as there are probably reviews (not to mention Wikipedia articles) that mention that a particular author is British.
As to awards "BSFA award winners" brings back some "interesting" suggestions
The Road, The Handmaid's Tale, The Windup Girl, Under the Skin, The Fifth Season all listed and all haven't won (or made the shortlist with the exception of The Windup Girl). It also include 4 films (none of which were from the few years when there was a film/media category). In fact its list of 11 items only contained 2 winners - The City And The City and Ancillary Justice
Trying with a specific year does a lot better "BSFA award 2018" brings back
Embers of War by Gareth Powell. The actual winner for novel for 2018 (award presented in 2019)
The Rift by Nina Allen. Winner for 2017 (presented in 2018 - so I will let GPT3 off there)
American War by Omar El Akkad. Never shortlisted for the BSFA.
Dreams Before the Start of Time by Anne Charnock. Shortlist for 2017
Exit West by Mohsin Hamed. Shortlist for 2017
"BSFA Award 2019" brings back
Embers of War by Gareth Powell. The actual winner for novel for 2018 (award presented in 2019)
Europe at Dawn by Dave Hutchison. Shortlist for 2018.
Revenant Gun by Yoon Ha Lee. Shortlist for 2018.
And then also includes The Loosening Skin and The Quantum Magician that have never been shortlisted in any year.
Sure I knew GPT-3's world ends in 2021 - but still five years before that was after 2014. I would have expected better results on the British part of the query as there are probably reviews (not to mention Wikipedia articles) that mention that a particular author is British.
As to awards "BSFA award winners" brings back some "interesting" suggestions
The Road, The Handmaid's Tale, The Windup Girl, Under the Skin, The Fifth Season all listed and all haven't won (or made the shortlist with the exception of The Windup Girl). It also include 4 films (none of which were from the few years when there was a film/media category). In fact its list of 11 items only contained 2 winners - The City And The City and Ancillary Justice
Trying with a specific year does a lot better "BSFA award 2018" brings back
Embers of War by Gareth Powell. The actual winner for novel for 2018 (award presented in 2019)
The Rift by Nina Allen. Winner for 2017 (presented in 2018 - so I will let GPT3 off there)
American War by Omar El Akkad. Never shortlisted for the BSFA.
Dreams Before the Start of Time by Anne Charnock. Shortlist for 2017
Exit West by Mohsin Hamed. Shortlist for 2017
"BSFA Award 2019" brings back
Embers of War by Gareth Powell. The actual winner for novel for 2018 (award presented in 2019)
Europe at Dawn by Dave Hutchison. Shortlist for 2018.
Revenant Gun by Yoon Ha Lee. Shortlist for 2018.
And then also includes The Loosening Skin and The Quantum Magician that have never been shortlisted in any year.
19timspalding
>18 andyl:
No, it's bad, I agree. LLMs (Large Language Models) like GPT3 are highly imperfect. Among other things, they make stuff up—something OpenAI calls "hallucination." It's a real problem. You can see some examples on a slide deck I made about this for a lightning talk yesterday ( https://www.slideshare.net/timspaldingperson/gpt3-api-vs-reality ). When asked to give me wife's books, for example, it comes up with the right one, but appends a description of each that is flat wrong. (The Mermaids Singing involves no actual Mermaids, and I assure you she's written no books whose protagonist is named "Boop.") It also invents whole books. For "Books about Jesuits in space," for example, it came up with Jesuits in Outer Space: Interstellar Voyages, Extraterrestrial Beings, and a Jesuitical Hermeneutic by Guy Consolmagno. This is believable, as Consolmagno, a Jesuit, is head of the Vatican observatory, has talked about the subgenre of "Jesuits in space" and written Would You Baptize an Extraterrestrial?. But he never wrote that book; it doesn't exist.
The solution we found was to ask it for books but not any accompanying text, to check everything against LibraryThing data, and only provide our versions of the data. This works a lot of the time, but it can include some wrong books. In its defense, however, other means are also error-prone. Google searches will have wrong things. A tag search on LibraryThing not infrequently includes books that are "tagged wrong," if that phrase even makes sense. Books get wrong BISAC codes all the time—publishers make mistakes, and they're incentivized to lie their books into categories to catch buyers—and even DDC and LCC have ratty data, intentional or not. My feeling is that search results are low-stakes enough that some errors are okay. But YMMV.
I think it might be interesting to see if we can't intercept some queries like you gave, and have LibraryThing say "I'm not sure, but I think you're looking for 2018 BAFTA winners. Here's a list from LibraryThing." World enough and time, we'd do that now. But I need to think if we can get GPT3 to return such metadata without a second query.
>11 Keeline: >15 SandraArdnas:
Yeah. First, I don't see anything inherently wrong in "AI." There is no intelligence; these systems are just doing very complex statistics. LibraryThing does a ton of traditional statistics already, including more AI-ish Bayesian stats that involves "training" models too complex to check, and now we do a small amount of neural-network stats. Recommendations and spam detection wouldn't exist without some mind-bendingly hard stats going on.
In none of this, however, do we create human content. We're not going to have computers write reviews, or tolerate if members do. We aren't against technology—look around. But we are against misrepresentation and the pollution of human air with AI smoke.
>12 klarusu:
I agree it's a potential issue. That said, it's already an issue on LT. Choose a tag like science fiction and you'll get a lot more white men than many would like—the residue of a lot of books bought in decades past, as well as the gap between contemporary aspirations and the statistics of the market.
As I've said before, we adjust for this slightly here and there already. Recommendations on LibraryThing have a logic that brings in a degree of gender parity. (e.g., if rank 1-10 are all men, rank 11, by a woman, will get promoted to 10 or 8 or 6; the same works for books that recommend mostly women). We do this across several other axes, like popularity and publication date, because changing things up is interesting. I will consider some rank-juggling in these results.
My sense is that GPT3 has a presentist bias that cuts against the usual lack of diversity in our data. But it's absolutely a live issue.
No, it's bad, I agree. LLMs (Large Language Models) like GPT3 are highly imperfect. Among other things, they make stuff up—something OpenAI calls "hallucination." It's a real problem. You can see some examples on a slide deck I made about this for a lightning talk yesterday ( https://www.slideshare.net/timspaldingperson/gpt3-api-vs-reality ). When asked to give me wife's books, for example, it comes up with the right one, but appends a description of each that is flat wrong. (The Mermaids Singing involves no actual Mermaids, and I assure you she's written no books whose protagonist is named "Boop.") It also invents whole books. For "Books about Jesuits in space," for example, it came up with Jesuits in Outer Space: Interstellar Voyages, Extraterrestrial Beings, and a Jesuitical Hermeneutic by Guy Consolmagno. This is believable, as Consolmagno, a Jesuit, is head of the Vatican observatory, has talked about the subgenre of "Jesuits in space" and written Would You Baptize an Extraterrestrial?. But he never wrote that book; it doesn't exist.
The solution we found was to ask it for books but not any accompanying text, to check everything against LibraryThing data, and only provide our versions of the data. This works a lot of the time, but it can include some wrong books. In its defense, however, other means are also error-prone. Google searches will have wrong things. A tag search on LibraryThing not infrequently includes books that are "tagged wrong," if that phrase even makes sense. Books get wrong BISAC codes all the time—publishers make mistakes, and they're incentivized to lie their books into categories to catch buyers—and even DDC and LCC have ratty data, intentional or not. My feeling is that search results are low-stakes enough that some errors are okay. But YMMV.
I think it might be interesting to see if we can't intercept some queries like you gave, and have LibraryThing say "I'm not sure, but I think you're looking for 2018 BAFTA winners. Here's a list from LibraryThing." World enough and time, we'd do that now. But I need to think if we can get GPT3 to return such metadata without a second query.
>11 Keeline: >15 SandraArdnas:
Yeah. First, I don't see anything inherently wrong in "AI." There is no intelligence; these systems are just doing very complex statistics. LibraryThing does a ton of traditional statistics already, including more AI-ish Bayesian stats that involves "training" models too complex to check, and now we do a small amount of neural-network stats. Recommendations and spam detection wouldn't exist without some mind-bendingly hard stats going on.
In none of this, however, do we create human content. We're not going to have computers write reviews, or tolerate if members do. We aren't against technology—look around. But we are against misrepresentation and the pollution of human air with AI smoke.
>12 klarusu:
I agree it's a potential issue. That said, it's already an issue on LT. Choose a tag like science fiction and you'll get a lot more white men than many would like—the residue of a lot of books bought in decades past, as well as the gap between contemporary aspirations and the statistics of the market.
As I've said before, we adjust for this slightly here and there already. Recommendations on LibraryThing have a logic that brings in a degree of gender parity. (e.g., if rank 1-10 are all men, rank 11, by a woman, will get promoted to 10 or 8 or 6; the same works for books that recommend mostly women). We do this across several other axes, like popularity and publication date, because changing things up is interesting. I will consider some rank-juggling in these results.
My sense is that GPT3 has a presentist bias that cuts against the usual lack of diversity in our data. But it's absolutely a live issue.
20Keeline
ChatGPT’s inability to deal with numbers is a frequent source of comment. Creating a prompt that uses it only highlights a known flaw in this language model.
James
James
22cpg
The AI Search "LibraryThing top books by rating" produces a list disjoint from the Zeitgeist Top Books by Rating list.
"Books not on LibraryThing" is pretty bad.
As is "Books unknown to GPT3".
"Books not on LibraryThing" is pretty bad.
As is "Books unknown to GPT3".
24paradoxosalpha
>13 reading_fox: Is this going to make all the treasure Hunts easier?!
Looks like it will. I pasted this clue into the search, and presto.
A circle of stone, and a fall through time,
Epic history and a love sublime,
This romance novel bestseller,
Is now a TV show most stellar.
It's just a "Works" search, though. So this author clue returned a dozen works, although the top ones were by the correct author.
Manners, courtship, social class,
Her famous work holds up a glass,
Examines these, and tells a tale,
Of love enacted in minute detail.
Looks like it will. I pasted this clue into the search, and presto.
A circle of stone, and a fall through time,
Epic history and a love sublime,
This romance novel bestseller,
Is now a TV show most stellar.
It's just a "Works" search, though. So this author clue returned a dozen works, although the top ones were by the correct author.
Manners, courtship, social class,
Her famous work holds up a glass,
Examines these, and tells a tale,
Of love enacted in minute detail.
25timspalding
>24 paradoxosalpha:
That's funny. When were playing as a company, we tried that. I found that it was only good maybe 1/3 of the time. Needless to say, we're going to make future hunts non-AI-able!
That's funny. When were playing as a company, we tried that. I found that it was only good maybe 1/3 of the time. Needless to say, we're going to make future hunts non-AI-able!
26timspalding
I made some improvements. It should return several more results, and be slightly higher quality.
27timspalding
Incidentally, it handles non-English queries without effort:
* Films français sur le passage à l'âge adulte
* Romanzo su un sottomarino sovietico canaglia
* Auctor Italicus qui carmen scripsit de Inferno adire
* Films français sur le passage à l'âge adulte
* Romanzo su un sottomarino sovietico canaglia
* Auctor Italicus qui carmen scripsit de Inferno adire
28anglemark
Almost any search I try yields dismal results, so I'll hold off from using this and perhaps try again after you've upgraded to the next version. Eventually, I think this will be a great addition to Search.
29thorold
>27 timspalding: Fun!
I tried “Schwuler Roman aus den Zwanziger Jahren” https://www.librarything.com/search.php?search=Schwuler+Roman+aus+den+Zwanziger+...
It picked up Mädchen in Uniform, Cabaret and Death in Venice, all at least vaguely on-topic even if none is actually a novel from the twenties, as well as Brecht’s play Man equals man and Klaus Theleweit’s Männerphantasien, which might sound as though they should be on-topic but clearly aren’t. But it totally failed to find the most obvious candidate, Klaus Mann’s Der fromme Tanz. Interesting that it does clearly focus on Germany for a query in German, anyway, which might be useful.
I tried “Schwuler Roman aus den Zwanziger Jahren” https://www.librarything.com/search.php?search=Schwuler+Roman+aus+den+Zwanziger+...
It picked up Mädchen in Uniform, Cabaret and Death in Venice, all at least vaguely on-topic even if none is actually a novel from the twenties, as well as Brecht’s play Man equals man and Klaus Theleweit’s Männerphantasien, which might sound as though they should be on-topic but clearly aren’t. But it totally failed to find the most obvious candidate, Klaus Mann’s Der fromme Tanz. Interesting that it does clearly focus on Germany for a query in German, anyway, which might be useful.
30Herenya
The search box doesn't show up on mobile (Chrome on Android). It shows if I switch to desktop mode.
31bnielsen
Nice feature!. Searching for "books written in two weeks" gave me stuff like
The Gambler by Fyodor Dostoevsky
Of course on LT "two weeks" could be anything :-)
The Gambler by Fyodor Dostoevsky
Of course on LT "two weeks" could be anything :-)
32conceptDawg
>30 Herenya: On mobile it uses the main search box at the top of the screen. Otherwise there would just be two search boxes on top of each other and that would be very confusing.
33SandraArdnas
Tried poetry about nature and while good in detecting major poets, it also pulled up a number of movies
34timspalding
Made it a little faster, better.
>33 SandraArdnas: Yeah, it's hard to get it to NOT provide movies and such, without totally excluding them, even when you want them.
>33 SandraArdnas: Yeah, it's hard to get it to NOT provide movies and such, without totally excluding them, even when you want them.
35SandraArdnas
>34 timspalding: I was more thinking about why it pulled those at all on the first page of results since I doubt any have anything to do with poetry, not so much about it differentiating media. It would make sense if the movie featured not only nature, but also poetry in some way. Perhaps it thinks the Call of the Wild movie is poetic, lol
36timspalding
Okay, if you ask for books, it will (mostly) only give you books.
Improving this, but the basics are there.
Improving this, but the basics are there.
38timspalding
Should work now. It won't work for everything, though. It's sensitive to sensitive questions.
39AnnieMod
>38 timspalding: Well, https://www.librarything.com/search.php?search=books+on+sensitive+topics&sea... :)
Sorry, could not resist. In the last years I often ask google with complete sentences - I am not sure how I would use that in LT but fuzzy searches are always a great thing - AI or no-AI being involved.
Sorry, could not resist. In the last years I often ask google with complete sentences - I am not sure how I would use that in LT but fuzzy searches are always a great thing - AI or no-AI being involved.
41timspalding
movies with very low rotten tomatoes scores
https://www.librarything.com/search.php?search=movies+with+very+low+rotten+tomat...
movies that bombed at the box office
https://www.librarything.com/search.php?search=movies+that+bombed+at+the+box+off...
https://www.librarything.com/search.php?search=movies+with+very+low+rotten+tomat...
movies that bombed at the box office
https://www.librarything.com/search.php?search=movies+that+bombed+at+the+box+off...
42timspalding
kids killing each other
https://www.librarything.com/search.php?search=kids+killing+each+other&searc...
https://www.librarything.com/search.php?search=kids+killing+each+other&searc...
43AntonioGallo
I like this idea. Please do carry on ...
44Authentico
As others have said, it doesn't work well with numbers. I tried 'books with 0 reviews' and all the books it gave me have more than 20 reviews (most of them over 200+).
https://www.librarything.com/search.php?search=books+with+0+reviews&searchty...
I'm not sure if that's a limit of the AI, or if there really isn't a way to see books with no reviews.
https://www.librarything.com/search.php?search=books+with+0+reviews&searchty...
I'm not sure if that's a limit of the AI, or if there really isn't a way to see books with no reviews.
45thorold
>44 Authentico: I don’t think ChatGPT is looking at the LT data, so it wouldn’t know about our reviews.
I don’t think it’s particularly good at negatives, either: https://www.librarything.com/search.php?search=Great+books+that+have+never+been+...
Trying to break it again with a completely illogical request: https://www.librarything.com/search.php?search=books+no-one+talks+about+on+the+i...
Predictably, the results don’t actually satisfy the search criterion, but it does give some well-known books that often get cited as “books you haven’t heard of”. This is the first “obscure books” AI search I’ve tried that didn’t return The shadow of the wind, at least!
Tim’s Latin Dante search from >27 timspalding: seems to be broken now.
I don’t think it’s particularly good at negatives, either: https://www.librarything.com/search.php?search=Great+books+that+have+never+been+...
Trying to break it again with a completely illogical request: https://www.librarything.com/search.php?search=books+no-one+talks+about+on+the+i...
Predictably, the results don’t actually satisfy the search criterion, but it does give some well-known books that often get cited as “books you haven’t heard of”. This is the first “obscure books” AI search I’ve tried that didn’t return The shadow of the wind, at least!
Tim’s Latin Dante search from >27 timspalding: seems to be broken now.
46timspalding
>44 Authentico: >45 thorold:
I think some LibraryThing data is in its huge dataset, acquired by scraping the web. But it doesn't have database access or any particular focus or interest in LibraryThing.
I think some LibraryThing data is in its huge dataset, acquired by scraping the web. But it doesn't have database access or any particular focus or interest in LibraryThing.
47Petroglyph
I like the crystal ball icon -- a subtle warning sign
48timspalding
If anyone is minded, I'd love to hear about queries you think are and aren't optimal. I'm currently working on varieties of queries to GPT, and also ways of turn queries into other types of searches.
For example, "science fiction about aliens in medieval times," ( https://www.librarything.com/search.php?search=science+fiction+about+aliens+in+m... ) is right to include The High Crusade and Doomsday Book. But it's missing Eifelheim and should not include anything from Outlander, which takes place in the 18th century, right?
I'm working on ways to make the native query to GPT give better results, also how to turn "science fiction about aliens in medieval times" into a tagmash of "aliens," "science fiction," "medieval," because if I can do that, I can use our tagmash page for that, which is the best https://www.librarything.com/tag/aliens%2C%20medieval%2C%20science%20fiction .
So, if you have feedback, let me know.
PS: Here's a naive example:
For example, "science fiction about aliens in medieval times," ( https://www.librarything.com/search.php?search=science+fiction+about+aliens+in+m... ) is right to include The High Crusade and Doomsday Book. But it's missing Eifelheim and should not include anything from Outlander, which takes place in the 18th century, right?
I'm working on ways to make the native query to GPT give better results, also how to turn "science fiction about aliens in medieval times" into a tagmash of "aliens," "science fiction," "medieval," because if I can do that, I can use our tagmash page for that, which is the best https://www.librarything.com/tag/aliens%2C%20medieval%2C%20science%20fiction .
So, if you have feedback, let me know.
PS: Here's a naive example:
Me: Give me several LibraryThing tagmashes that would correspond to the phrase "science fiction about aliens in medieval times"
GPT3.5: Here are several different LibraryThing tagmashes that correspond to the phrase "science fiction about aliens in medieval times":
"science fiction, aliens, medieval, time travel"
"science fiction, aliens, medieval, alternate history"
"science fiction, aliens, medieval, historical fantasy"
"science fiction, aliens, medieval, space opera"
"science fiction, aliens, medieval, first contact"
"science fiction, aliens, medieval, parallel universes"
"science fiction, aliens, medieval, epic fantasy"
"science fiction, aliens, medieval, sword and sorcery"
"science fiction, aliens, medieval, interdimensional travel"
"science fiction, aliens, medieval, dark fantasy"
49andyl
>48 timspalding:
Yeah there is plenty not to like about the "science fiction about aliens in medieval times," - Wells for instance and "Star Trek: The Motion Picture" the novelization.
But also this is one case where LT's tagmash (aliens, medieval, science fiction) isn't great either
It brings back Anathem (set in the far future) based on I think 1 person tagging it medieval. Same for A Fire Upon The Deep and Saga vol. 1 and Neverness and The Steerswoman and probably more. Even The Colour of Magic (humorous fantasy set in a made-up universe) gets on the list with just 2 people tagging as "medieval" in the 20634 owners.
Yeah there is plenty not to like about the "science fiction about aliens in medieval times," - Wells for instance and "Star Trek: The Motion Picture" the novelization.
But also this is one case where LT's tagmash (aliens, medieval, science fiction) isn't great either
It brings back Anathem (set in the far future) based on I think 1 person tagging it medieval. Same for A Fire Upon The Deep and Saga vol. 1 and Neverness and The Steerswoman and probably more. Even The Colour of Magic (humorous fantasy set in a made-up universe) gets on the list with just 2 people tagging as "medieval" in the 20634 owners.
50Nick-Myra
First, I think that AI-Search is a very good idea - well done on trialling it.
I support using GPT to assist in the search, in the 19 hits it found on my search, only 1 was a lemon - I can live with that.
I might add, that I found LibraryThing because I used Google to find the title of a book using only the first line, and based on that I joined up here. I have been very impressed with the site - thanks to developers and all the active members.
Later on I searched for something slightly more detailed (content-wise) in the same book, and a smaller search engine was the only one that found it - not one of the big 3 SE's.
To test LT.GPT.search I put it through its paces as follows :
I watched fragments of a movie on TV 3 years ago, I have searched and searched through all chess movies to find it (47 in the list at last count) but have not found the actual movie yet. So it's an important search for me.
Here is the top level gist of its plot, my search string
"two sisters playing chess, separated for many years, find each other later to finish their game"
I repeated this search twice with the identical terms, about 5 minutes apart.
This gave quite different results - that is why I am posting my finding here
@# 1st search #14 total, 6 unique to 1st search
@# 2nd identical search terms #10 total, 5 unique to 2nd search
Edit: something wrong with that 14 (6 unq) and 10 (5 unq) split, since the Venn set is not consistent { 6 {8 or 5} 5 } !
I think I miscounted the first 6 unique, I am fairly sure of the 2nd 5 unq (and I'm not going to manually count again) - lets assume it is
Venn: {9 {5} 5} --> so that the nonrepeatability in the search is something like 14/19 or 74%
Also it seemed to pick up only books that explicitly mention "chess" and not "sisters". Does the search only consider the book title and the summary, not member reviews or member comments?
So a suggestion, have search parameters that allow the search scope to be tunable.
Then it found this book below, I am yet to understand why ...
{why this ?}
no mention of "chess" or "sister" at all in the title, summary, all reviews and comments ?
The Collected Stories of Eudora Welty
by Eudora Welty
https://www.librarything.com/work/34980
This omnibus volume by one of the South's greatest writers includes stories published prior to 1980. Stories are as good in themselves and as influential on the aspirations of others as any since Hemingway's. The breadth of Welty's offering is finally most visible not in the variety of types--farce, satire, horror, lyric, pastoral, mystery--but in the clarity and solidity and absolute honesty of a lifetime's vision.
{/why }
I support using GPT to assist in the search, in the 19 hits it found on my search, only 1 was a lemon - I can live with that.
I might add, that I found LibraryThing because I used Google to find the title of a book using only the first line, and based on that I joined up here. I have been very impressed with the site - thanks to developers and all the active members.
Later on I searched for something slightly more detailed (content-wise) in the same book, and a smaller search engine was the only one that found it - not one of the big 3 SE's.
To test LT.GPT.search I put it through its paces as follows :
I watched fragments of a movie on TV 3 years ago, I have searched and searched through all chess movies to find it (47 in the list at last count) but have not found the actual movie yet. So it's an important search for me.
Here is the top level gist of its plot, my search string
"two sisters playing chess, separated for many years, find each other later to finish their game"
I repeated this search twice with the identical terms, about 5 minutes apart.
This gave quite different results - that is why I am posting my finding here
@# 1st search #14 total, 6 unique to 1st search
@# 2nd identical search terms #10 total, 5 unique to 2nd search
Edit: something wrong with that 14 (6 unq) and 10 (5 unq) split, since the Venn set is not consistent { 6 {8 or 5} 5 } !
I think I miscounted the first 6 unique, I am fairly sure of the 2nd 5 unq (and I'm not going to manually count again) - lets assume it is
Venn: {9 {5} 5} --> so that the nonrepeatability in the search is something like 14/19 or 74%
Also it seemed to pick up only books that explicitly mention "chess" and not "sisters". Does the search only consider the book title and the summary, not member reviews or member comments?
So a suggestion, have search parameters that allow the search scope to be tunable.
Then it found this book below, I am yet to understand why ...
{why this ?}
no mention of "chess" or "sister" at all in the title, summary, all reviews and comments ?
The Collected Stories of Eudora Welty
by Eudora Welty
https://www.librarything.com/work/34980
This omnibus volume by one of the South's greatest writers includes stories published prior to 1980. Stories are as good in themselves and as influential on the aspirations of others as any since Hemingway's. The breadth of Welty's offering is finally most visible not in the variety of types--farce, satire, horror, lyric, pastoral, mystery--but in the clarity and solidity and absolute honesty of a lifetime's vision.
{/why }
51Stevil2001
I've been messing with it a bit, and most of the time, to be honest, it is not great: misses stuff regular search would bring up, comes up with stuff that doesn't make sense.
But I just did one it did surprisingly well with: https://www.librarything.com/search.php?search=oz+books+by+royal+historians+othe...
It correctly identifies who a "royal historian" is (none of the Oz continuations by non-royal historian authors are included), includes ones by royal historians even if they weren't officially part of the "Famous Forty," and doesn't include any by Baum with one exception.
EDIT: Slightly different and slightly better results if you capitalize "Baum," oddly enough: https://www.librarything.com/search.php?search=oz+books+by+royal+historians+othe...
But I just did one it did surprisingly well with: https://www.librarything.com/search.php?search=oz+books+by+royal+historians+othe...
It correctly identifies who a "royal historian" is (none of the Oz continuations by non-royal historian authors are included), includes ones by royal historians even if they weren't officially part of the "Famous Forty," and doesn't include any by Baum with one exception.
EDIT: Slightly different and slightly better results if you capitalize "Baum," oddly enough: https://www.librarything.com/search.php?search=oz+books+by+royal+historians+othe...
52prosfilaes
>50 Nick-Myra: Also it seemed to pick up only books that explicitly mention "chess" and not "sisters". Does the search only consider the book title and the summary, not member reviews or member comments?
It doesn't do any of that. Chat GPT crawled some large chunk of the Internet and transformed that into a text producing machine. It's unknown how much of LT it processed, but it certainly processed huge chunks beyond LT including the text of many books. It's possible it's actually read this book. It's also possible that it's coming up with the name out of thin air; these type of text generators are known to "hallucinate" and to answer why to any specific question may be impossible, or solvable by a year of researcher time armed with a supercomputer. These systems are built through processes that make any specific why hard to answer, even for those intimately familiar with the system.
It doesn't do any of that. Chat GPT crawled some large chunk of the Internet and transformed that into a text producing machine. It's unknown how much of LT it processed, but it certainly processed huge chunks beyond LT including the text of many books. It's possible it's actually read this book. It's also possible that it's coming up with the name out of thin air; these type of text generators are known to "hallucinate" and to answer why to any specific question may be impossible, or solvable by a year of researcher time armed with a supercomputer. These systems are built through processes that make any specific why hard to answer, even for those intimately familiar with the system.
53prosfilaes
>48 timspalding: should not include anything from Outlander, which takes place in the 18th century, right?
Technically, but https://www.librarything.com/tag/18th%20century%2C%20medieval offers us Boswell and Swift, Outlander and The Dream of the Red Chamber. Even the academics get in on it, with the at best carelessly titled The Jew in the Medieval World: A Source Book, 315-1791 and Medieval Persia, 1040–1797.
And The Heavenly City of the Eighteenth-Century Philosophers has a description that says "Carl Becker demonstrates that Voltaire, Hume, Diderot, and Locke were living in a medieval world...".
Technically, but https://www.librarything.com/tag/18th%20century%2C%20medieval offers us Boswell and Swift, Outlander and The Dream of the Red Chamber. Even the academics get in on it, with the at best carelessly titled The Jew in the Medieval World: A Source Book, 315-1791 and Medieval Persia, 1040–1797.
And The Heavenly City of the Eighteenth-Century Philosophers has a description that says "Carl Becker demonstrates that Voltaire, Hume, Diderot, and Locke were living in a medieval world...".
54timspalding
>51 Stevil2001:
There's going to be another version soon. I look forward to hearing what people have to thin.
There's going to be another version soon. I look forward to hearing what people have to thin.
55jjwilson61
>53 prosfilaes: Medieval is a term that depends on the region. Medieval Europe is going to be a different time period than Medieval Persia.
56Nick-Myra
>52 prosfilaes: Chat GPT crawled some large chunk of the Internet and transformed that into a text producing machine.
Curiously enough I had a good worksession with CGPT last night, put it through its paces summarising two books, querying it about a defamation case looming against it, gave it a maths proof and a code assignment, managed to coerce it to play the first few moves in a game of chess (it tried to avoid playing), and then some banter about LaMDA - eventually I gave it a request that crashed it.
So to your response, in response to answering some questions about famous chess games it gave this response
assistant> I'm sorry, as an AI language model, I am not able to access external websites or their contents. However, it's possible that the notes accompanying the game on chess.com might reference a similar game that inspired the author's depiction
52> These systems are built through processes that make any specific why hard to answer, even for those intimately familiar with the system.
Yes indeed, I amended my first post to look at the degree of nonrepeatability :
50> Venn: 9 5 5 --> so that the nonrepeatability in the search is something like 14/19 or 74%
I know that CGPT has a dispersion parameter, set quite aggressively, but it is not confidence inspiring to have so much inconsistency.
Curiously enough I had a good worksession with CGPT last night, put it through its paces summarising two books, querying it about a defamation case looming against it, gave it a maths proof and a code assignment, managed to coerce it to play the first few moves in a game of chess (it tried to avoid playing), and then some banter about LaMDA - eventually I gave it a request that crashed it.
So to your response, in response to answering some questions about famous chess games it gave this response
assistant> I'm sorry, as an AI language model, I am not able to access external websites or their contents. However, it's possible that the notes accompanying the game on chess.com might reference a similar game that inspired the author's depiction
52> These systems are built through processes that make any specific why hard to answer, even for those intimately familiar with the system.
Yes indeed, I amended my first post to look at the degree of nonrepeatability :
50> Venn: 9 5 5 --> so that the nonrepeatability in the search is something like 14/19 or 74%
I know that CGPT has a dispersion parameter, set quite aggressively, but it is not confidence inspiring to have so much inconsistency.
57Doondeck
>56 Nick-Myra: How do you navigate to it?
59melannen
I'm also an AI skeptic, but I agree that interfaces to actual database searches is one of the best use cases for a natural language model! This is really interesting, and I look forward to seeing ways it interfaces even better with LT's data.
As you'd expect though, as well as not knowing about things published post-2021, it's also not great with stuff that hasn't been discussed a lot on the internet in the past fifteen years or so; I can't get any variation on "science fiction about Jesuits" to give me anything about The Order of Saint Vidicon, for example! (Tagmash gets there right away.)
When I try to get my old childrens' book about dollhouse dolls come to life and fighting crime that's my usual stumper for book searches, it's pulled up all the usual wrong suggestions, but not the correct answer that Name That Book has found, The Dollhouse Caper.
It is interesting, though, that unlike most of the search engines I'm used to, it tends to give me *more* results the longer and more detailed my query is.
As you'd expect though, as well as not knowing about things published post-2021, it's also not great with stuff that hasn't been discussed a lot on the internet in the past fifteen years or so; I can't get any variation on "science fiction about Jesuits" to give me anything about The Order of Saint Vidicon, for example! (Tagmash gets there right away.)
When I try to get my old childrens' book about dollhouse dolls come to life and fighting crime that's my usual stumper for book searches, it's pulled up all the usual wrong suggestions, but not the correct answer that Name That Book has found, The Dollhouse Caper.
It is interesting, though, that unlike most of the search engines I'm used to, it tends to give me *more* results the longer and more detailed my query is.
60bnielsen
I'm currently reading a book about the Claas Relotius scandal. "Tausend Zeilen Lüge. Das System Relotius und der deutsche Journalismus."
In short Relotius for years invented stories and interviews without researching anything in real life. As I read the book I keep thinking of him as some sort of human version of chatGPT. It writes what you'd like to read.
In short Relotius for years invented stories and interviews without researching anything in real life. As I read the book I keep thinking of him as some sort of human version of chatGPT. It writes what you'd like to read.
61prosfilaes
>56 Nick-Myra: "assistant> I'm sorry, as an AI language model, I am not able to access external websites or their contents. However, it's possible that the notes accompanying the game on chess.com might reference a similar game that inspired the author's depiction"
Right. It doesn't have current access to the Internet. But the data that created it was compiled from all the textual data they could find, including huge Internet text crawls.
Right. It doesn't have current access to the Internet. But the data that created it was compiled from all the textual data they could find, including huge Internet text crawls.
62prosfilaes
>55 jjwilson61: Definitions of medieval I found all point to about 500-1500 Europe, some pointing instead o Middle Ages, then defined the same way. It's sometimes stretched to the world. I see no standards for differing time periods; e.g. Keith Knapp says "China specialists have long debated the applicability of this patently Western concept to Chinese history. Some deny its utility outright. Many other scholars, though, assume that China had a medieval period. Yet amongst them, no consensus exists over when it started, how long it lasted, or what were its characteristics. Since these issues have proved intractable, Sinologists nowadays largely ignore them, as if they were unimportant." and then argues for 200-1000 CE. The "Medieval Persia" chapter in The Oxford Handbook of Iranian History covers 680-1200 CE.
Certainly The Jew in the Medieval World: A Source Book, 315-1791, ending with the partition of Poland, is stretching medieval by any definition, as does "Voltaire, Hume, Diderot, and Locke were living in a medieval world...".
Certainly The Jew in the Medieval World: A Source Book, 315-1791, ending with the partition of Poland, is stretching medieval by any definition, as does "Voltaire, Hume, Diderot, and Locke were living in a medieval world...".
63paradoxosalpha
>62 prosfilaes:
I can imagine that The Jew in the Medieval World: A Source Book, 315-1791 might include sources from later centuries still trained on the medieval period. The fourth century C.E. is typically viewed as Late Antiquity rather than the Middle Ages, though.
"Voltaire, Hume, Diderot, and Locke were living in a medieval world"? No. No way. No how. Their modernity wasn't even "Early"!
I can imagine that The Jew in the Medieval World: A Source Book, 315-1791 might include sources from later centuries still trained on the medieval period. The fourth century C.E. is typically viewed as Late Antiquity rather than the Middle Ages, though.
"Voltaire, Hume, Diderot, and Locke were living in a medieval world"? No. No way. No how. Their modernity wasn't even "Early"!
64Aquila
>25 timspalding: Just tested the current (poetry) hunt - could get all 5 clues that led to work or author pages with it.
65Taliesien
An external site search of LT is useful as well. There were a few clues that I knew but didn't have the foggiest idea how to get there from within LT.
66timspalding
>65 Taliesien:
We had an internal debate about this. I lost. I would have made sure all were tested and impossible. Alas.
We had an internal debate about this. I lost. I would have made sure all were tested and impossible. Alas.
67timspalding
Did you use ChatGPT or the AI search?
68timspalding
Shit. I just tried it. Yup. Proves how good AI search is, but I hate it. I'm going to win the next argument.
69Aquila
I initially cherry picked clauses to use for the AI search but for most of them it turned out you could just wang the whole clue in and get the answer. I think 8 worked better when finessed a bit.
70paradoxosalpha
I think I finished this hunt in record time--before I even thought to use AI search.
71Aquila
Yeah, it didn't occur to me until partway through, so I waited until the end and then ran them.
72SandraArdnas
>67 timspalding: But we provide clues to one another all the time, does it really matter if AI search can help? It's fun, but not cut-throat competition :D
73paradoxosalpha
It does sort of trivialize those particular clues (ones with book or author page solutions), which in this case were less than half the hunt.
75ablachly
>68 timspalding: I'm on the side of it not mattering. Some people will google for the answer. Some people will use clues. Some people would never do either. If someone chooses to use the AI Search, then that's fine?
76paradoxosalpha
Maybe there could be a notice provided to this effect:
These hunts are just for fun and to learn about LT features. The AI search makes some of these clues very easy, and you may wish to abstain from it if you enjoy puzzling out the clues. Some people also avoid Google and other outside search engines, preferring the challenge to find the answers entirely within LT.
These hunts are just for fun and to learn about LT features. The AI search makes some of these clues very easy, and you may wish to abstain from it if you enjoy puzzling out the clues. Some people also avoid Google and other outside search engines, preferring the challenge to find the answers entirely within LT.
77Stevil2001
Part of the point of the hunt is to introduce aspects of LibraryThing, right? It's a good introduction to a very new aspect of LT.
78Aquila
Honestly I don't think it needs any "some people also avoid Google" comment. Everyone has different general knowledge, so obvious things for some people are very inobvious for others, especially if they aren't English as a first language Americans who live in Maine, no need to shame people for the way they find sufficient information to answer the clue.
79paradoxosalpha
>78 Aquila:
Well, no shame was intended, especially since I'm not one who eschews Google. It was just an observation derived from the hunt threads that I thought was germane.
Well, no shame was intended, especially since I'm not one who eschews Google. It was just an observation derived from the hunt threads that I thought was germane.
80timspalding
See https://www.librarything.com/topic/350287 . I've relaunched the feature with some new features in it.
81Keeline
The "some people also avoid Google" statement reminds me of the parlor game my wife and I would play when we were watching the PBS show called History Detectives. There would be some object that a person wanted to learn something about. One of their hosts / historians would do some research in various places and report back to the person. A given investigation often used the phrase "and what s/he told me next..." for some part that we don't hear but is revealed by the host / historian in the wrap-up segment.
While watching a show we would try to research the item in question and could usually unravel its history quicker than the show could reveal it. Our search resources included newspaper archives (to which we subscribe to four), genealogy databases, and patent searches. We are good with these tools and that was a key to our success.
I can imagine the support staff that might be employed on a show like this and the producers to arrange trips and visits to resources so the host / historian could "discover" something.
The one thing we tried to avoid was a straight Google search because this could easily bring up a transcript of the show and that felt too much like cheating.
James
While watching a show we would try to research the item in question and could usually unravel its history quicker than the show could reveal it. Our search resources included newspaper archives (to which we subscribe to four), genealogy databases, and patent searches. We are good with these tools and that was a key to our success.
I can imagine the support staff that might be employed on a show like this and the producers to arrange trips and visits to resources so the host / historian could "discover" something.
The one thing we tried to avoid was a straight Google search because this could easily bring up a transcript of the show and that felt too much like cheating.
James
82JuliaMaria
Great feature! Needed when you forgot the name or author of a novel.
I would love to have it in German as well.
I did not find all books I was looking for based on content but some were easy to find.
Would also be nice to restrict it to your own library.
I would love to have it in German as well.
I did not find all books I was looking for based on content but some were easy to find.
Would also be nice to restrict it to your own library.
83kristilabrie
>82 JuliaMaria: I believe it should know other languages (I tested some French searches, at least): does it not? It's also available on the German website at https://www.librarything.de/search.php?searchtype=ai, if that's what you're looking for.
Do share some searches that you're finding it's struggling with, we can take a closer look!
Do share some searches that you're finding it's struggling with, we can take a closer look!
85thorold
Trying a few German prompts (books by a Romanian-German writer, and books about Leipzig):
https://www.librarything.com/search.php?search=Deutsche+Schriftstellerin+aus+Rum... gives no results, but https://www.librarything.com/search.php?search=Romane+von+einer+deutschen+Schrif... works correctly, but it only gives the most obvious three results.
https://www.librarything.com/search.php?search=Bücher+über+die+Stadt+Leipzig&a... Fails, https://www.librarything.com/search.php?search=Bücher+über+die+Geschichte+der+... gives results that are a mixture of useful and nonsense, whilst
https://www.librarything.com/search.php?search=Romane%2C+die+in+Leipzig+spielen&... gives results, but none of them really meet the search criteria
https://www.librarything.com/search.php?search=Deutsche+Schriftstellerin+aus+Rum... gives no results, but https://www.librarything.com/search.php?search=Romane+von+einer+deutschen+Schrif... works correctly, but it only gives the most obvious three results.
https://www.librarything.com/search.php?search=Bücher+über+die+Stadt+Leipzig&a... Fails, https://www.librarything.com/search.php?search=Bücher+über+die+Geschichte+der+... gives results that are a mixture of useful and nonsense, whilst
https://www.librarything.com/search.php?search=Romane%2C+die+in+Leipzig+spielen&... gives results, but none of them really meet the search criteria
86kristilabrie
>85 thorold: Testing your searches, since the AI Search may produce different results for the same searches:
https://www.librarything.com/search.php?search=Deutsche+Schriftstellerin+aus+Rum... gives me 1 result (The Passport by Herta Müller), https://www.librarything.com/search.php?search=Romane+von+einer+deutschen+Schrif... gives me 15 results.
https://www.librarything.com/search.php?search=B%C3%BCcher+%C3%BCber+die+Stadt+L... gives me 1 book about Leipzig architecture, https://www.librarything.com/search.php?search=B%C3%BCcher+%C3%BCber+die+Geschic... (books about the history of the city of Leipzig) gives results that appear to be a mix of useful and nonsense (so, same as yours), and https://www.librarything.com/search.php?search=Romane%2C+die+in+Leipzig+spielen&... (novels set in Leipzig) I checked a handful and most had some setting in Leipzig but not all that I could easily find (though they could have been).
Passing this info to timspalding if it's helpful.
https://www.librarything.com/search.php?search=Deutsche+Schriftstellerin+aus+Rum... gives me 1 result (The Passport by Herta Müller), https://www.librarything.com/search.php?search=Romane+von+einer+deutschen+Schrif... gives me 15 results.
https://www.librarything.com/search.php?search=B%C3%BCcher+%C3%BCber+die+Stadt+L... gives me 1 book about Leipzig architecture, https://www.librarything.com/search.php?search=B%C3%BCcher+%C3%BCber+die+Geschic... (books about the history of the city of Leipzig) gives results that appear to be a mix of useful and nonsense (so, same as yours), and https://www.librarything.com/search.php?search=Romane%2C+die+in+Leipzig+spielen&... (novels set in Leipzig) I checked a handful and most had some setting in Leipzig but not all that I could easily find (though they could have been).
Passing this info to timspalding if it's helpful.
87timspalding
>85 thorold:
All of the ones that failed work now. I think we're seeing ChatGPT's AI fail a lot. I'll add in some logic on that.
All of the ones that failed work now. I think we're seeing ChatGPT's AI fail a lot. I'll add in some logic on that.
88lorax
timspalding (#68):
Who cares? It's like cheating at solitaire. So they get some pixels they didn't "earn". People have been googling clues forever.
Who cares? It's like cheating at solitaire. So they get some pixels they didn't "earn". People have been googling clues forever.
89proximity1
"The Cadre in the Code" | How artificial intelligence could supplement and reinforce our emerging thought police | by Robert Henderson | (Spring, 2023 Issue: City Journal (New York)
--------------------------------
... "Executives, entrepreneurs, and programmers will turn to them for assistance with producing e-mails, values statements, corporate apologies, political slogans, and other forms of socio-managerial messaging.
"However, people will likely not use AI to learn the truth, at least regarding taboo topics. Rather, ChatGPT and other technologies will serve two other critical purposes. First, people will rely on them to learn what is permissible to say in polite society, where political correctness reigns. A Cato/YouGov survey found that while only 25 percent of those with a high school diploma or less regularly self-censor their political opinions, the figure reaches an astounding 44 percent among people with graduate degrees. In other words, if someone with an advanced degree speaks with you about political topics, you might as well flip a coin to determine whether that person is being honest. Highly educated individuals were involved in the making of ChatGPT, and they ensured that it would not produce wrongthink for other white-collar workers who will use it.
"Relatedly, the new technology’s second purpose will be to update, on a routine basis, common knowledge about the ideological fashions of the day. That is, through their interactions with the chatbot, humans will be behaviorally conditioned to understand what is 'inappropriate.'
"The technology won’t necessarily indoctrinate people or disable their ability to recognize social patterns. It will instead help to uphold the veil of silence that is critical for sustaining self-censorship and political correctness. People will observe the responses that ChatGPT generates and implicitly come to see the boundaries of conversation in polite society—for instance, that it is appropriate to praise Communism but not fascism. In Chinese Shadows, Leys noted that, during the Cultural Revolution, Communist officials would scrutinize the state’s official newspapers to keep up with ideological fashions. In the U.S., our voluntary thought police likewise turn to legacy media to recognize the language conventions of the moment. ChatGPT will supplement this purpose.
"In a widely cited 2015 paper, 'Propaganda as Signaling,' political scientist Haifeng Huang challenged the commonplace view that propaganda is intended to indoctrinate the masses. Indeed, propaganda is often preposterous and unpersuasive. Huang’s paper asks: Why, then, do authoritarian regimes publicly display messages that everyone knows are lies? ...
LT, obviously, has already chosen to implement ChatGPT. This was a decision taken without bothering to include LT's members in any decisive manner.
No one adopts ChatGPT for avowed purposes which are or are likely to be seen as noxiously authoritarian in character. It's never going to be proposed as a useful means to find, filter and remove "good" things. Everyone's intentions and purposes for the use of ChatGPT are always and above all guided by motives which are based purely on sweetness and light--never anything vicious, foul, censorious and, least of all, fucking evil.
But ChatGPT lends itself just as readily to what are very reasonably described as falling into what are vicious, foul, censorious and plain fucking evil ends. If you expect these to be openly admitted, you are adopting a child-like naive approach to a very seriously dangerous technology--one which shall be used, tweaked and operated from behind an opaque veil. You, ordinary reader, began and shall be and shall remain on the "ignorant" side of that veil.
90proximity1
As long ago as the mid-1960s, computer scientists had better, clearer grasps of what a computer was and was not, what it could do and could not do. These men and women did not foolishly anthropomorphize computers as mechanical versions of human brains. They did not misrepresent their machine work as "thinking" and, today, they would scoff and the concept and the use of the phrase, "artificial intelligence. This was so well grasped that these understandings made their way into popular culture's dramas--just as today, we represent the technology of the present and what is expected from it in the near future, in the 1960s' popular televsion programs, computers were sometimes a feature of plot-line.
A case example is the television series, "The Fugitive", which starred David Janssen as the character Dr. Richard Kimble, M.D., an escaped condemned man, on the run after having been wrongly convicted of having murdered his wife.
In episode 27 of season 3, entitled, "2130", after the large room-size computer-- something like the Univacs then in use at universities and government research centers-- a computer is put to use to try and find patterns of movement which could help the authorities predict the fugitive's most probable next movements--and so catch up with him.
Here, at the link, is an instructive excerpt from the television episode, cued up to the story, already in progress:
notice the careful use of terminology and the way that a computer expert corrects police detective Gerrard's mistaken assumption about the nature of the computer's use in prediction of the fugitive's movements:
... "it doesn't 'think', it computes."
"Play" > from here (12m:30secs into the program to 14:45, or two minutes and 15 seconds of play.)
https://youtu.be/hTBW2JHjGRs?t=750
Today, almost 60 years later, we've adopted foolish notions about "artificial 'intelligence'" and we're considerably less intelligent about it than were the people who put this drama together--and the computer experts they consulted.
A case example is the television series, "The Fugitive", which starred David Janssen as the character Dr. Richard Kimble, M.D., an escaped condemned man, on the run after having been wrongly convicted of having murdered his wife.
In episode 27 of season 3, entitled, "2130", after the large room-size computer-- something like the Univacs then in use at universities and government research centers-- a computer is put to use to try and find patterns of movement which could help the authorities predict the fugitive's most probable next movements--and so catch up with him.
Here, at the link, is an instructive excerpt from the television episode, cued up to the story, already in progress:
notice the careful use of terminology and the way that a computer expert corrects police detective Gerrard's mistaken assumption about the nature of the computer's use in prediction of the fugitive's movements:
... "it doesn't 'think', it computes."
"Play" > from here (12m:30secs into the program to 14:45, or two minutes and 15 seconds of play.)
https://youtu.be/hTBW2JHjGRs?t=750
Today, almost 60 years later, we've adopted foolish notions about "artificial 'intelligence'" and we're considerably less intelligent about it than were the people who put this drama together--and the computer experts they consulted.
91prosfilaes
>90 proximity1: As long ago as the mid-1960s, ... they would scoff and the concept and the use of the phrase, "artificial intelligence.
Alice M. Pierce's A Concise Bibliography of the Literature on Artificial Intelligence came out in 1959. And given that one of the biggest computers you could buy in 1965 had a 512 KB of memory and ran at 175 kIPS (i.e. at least 10,000 times slower than any modern cellphone), the beast was a bit of a different thing.
Alice M. Pierce's A Concise Bibliography of the Literature on Artificial Intelligence came out in 1959. And given that one of the biggest computers you could buy in 1965 had a 512 KB of memory and ran at 175 kIPS (i.e. at least 10,000 times slower than any modern cellphone), the beast was a bit of a different thing.
93bacchus.
>48 timspalding: How about vectorizing the tag mashes?
95HenrySt123
While reading about Jack London this morning, I became curious about the circle of writers, photographers, and others he associated with in San Francisco and Carmel. Particularly a man sometimes referred to as his mentor, George Sterling (Russ Brissenden in Martin Eden). Curious to learn more. Apparently, there’s a chapter on the group in Albert Parry’s Garrets and Pretenders: A History of Bohemianism in America (1933), but it seems there’s material for an entire book there.
Tried LT AI with the query, “what book tells the story of bohemian circle at carmel circa 1900.” The search returned 12 hits, 9 discrete titles (three books were listed twice). All had “Bohemian” in the title. Two were about Bohemians in Paris. One is about Bohemian Grove (geographically closer, but still not what I was looking for). Two looked as if they might be relevant: a novel based on the life of Dorothea Lange, and a study about Mark Twain, Bret Harte, and other San Francisco-based writers.
Rated the result three stars.
Then I altered the query to "what book tells the story of bohemian circle involving George Sterling and Jack London."
This yielded fewer results (six). Four came up in the first query, one new title was about Brooklyn, another about Jack London.
I don't think well-informed librarians have to fear for their jobs yet (at least, not because of AI).
Tried LT AI with the query, “what book tells the story of bohemian circle at carmel circa 1900.” The search returned 12 hits, 9 discrete titles (three books were listed twice). All had “Bohemian” in the title. Two were about Bohemians in Paris. One is about Bohemian Grove (geographically closer, but still not what I was looking for). Two looked as if they might be relevant: a novel based on the life of Dorothea Lange, and a study about Mark Twain, Bret Harte, and other San Francisco-based writers.
Rated the result three stars.
Then I altered the query to "what book tells the story of bohemian circle involving George Sterling and Jack London."
This yielded fewer results (six). Four came up in the first query, one new title was about Brooklyn, another about Jack London.
I don't think well-informed librarians have to fear for their jobs yet (at least, not because of AI).
96proximity1
"I don't think well-informed librarians have to fear for their jobs yet (at least, not because of AI)."
Apparently, that's based on an assumption that, because well-informed and competent librarians are vastly superior to some shoddy algorithmic-based search program, they shall not be at risk of losing their jobs to that inferior program.
Or, to put it in other words, the idea seems to be that crap technology just doesn't threaten and replace superior human actors. My personal experience every single day defies and confounds such an assumption.
A.I. truly sucks; and it shall replace--and already has replaced-- many superior human agents. Vicious stupidity is an exremely powerful force. A.I. is exclusively a backward-looking technology. It has only yesterday's data at best.
Imagine the game of "telephone"-- where a message is passed from one to another in a chain. Now, combine that method with a rear-view-only mirrored view of the world-- now, taking these together, suppose we assign the "resulting" technique the task of driving a motor car. Rear-view only, all other data telephoned through a series of communicants.
What you get is a "driver" which cannot adequately cope with any conditions which have not been received and filtered through the preliminaries. Therefore, a road closed, or a major construction pit right in the middle of the traffic lane or a bridge out-- any of these or hundreds of other circumstances arising in the previous minute-- means the A.I. "auto-pilot" plows right ahead, oblivious to the changed conditions.
97DugsBooks
I stumbled into this topic while searching for recently published books etc. on A.I. and could not find first published dates easily on the books listed. I was looking for the recently published.
98kristilabrie
>97 DugsBooks: From >1 timspalding:: "There are some other notes on the pop-up when you enter. Notably, it can't find most books published after September 2021."
It's best at finding books before September of 2021, I suspect that's where you were seeing issues?
It's best at finding books before September of 2021, I suspect that's where you were seeing issues?
99melannen
>96 proximity1: AI won't replace librarians anytime soon because most of a librarian's current reference job is showing people how to get to google and what to type into it and how to interpret the results they get. (most librarians aren't any better at it than your average LT user, which is still about 200x better than your average library patron, who probably doesn't know what a web browser or search engine is.)
Worst case scenario is librarians have to pivot to showing people how to get to chat search and what to type into it and how to interpret the results.
Worst case scenario is librarians have to pivot to showing people how to get to chat search and what to type into it and how to interpret the results.
100paradoxosalpha
>99 melannen:
I think that is fair with respect to public library workers, as opposed to librarians in the professional sense, who deal with acquisitions, cataloging, deaccession, and the many other tasks that go into creating and maintaining library collections. AI is likely to enlisted in these tasks as well, and "efficiencies" there may actually contribute to slimming library staff.
I think that is fair with respect to public library workers, as opposed to librarians in the professional sense, who deal with acquisitions, cataloging, deaccession, and the many other tasks that go into creating and maintaining library collections. AI is likely to enlisted in these tasks as well, and "efficiencies" there may actually contribute to slimming library staff.
101AntonioGallo
reality vs fiction https://www.librarything.com/search.php?search=reality+vs+fiction&searchtype... reality wins
"Reality refers to the state of things as they actually exist, while fiction refers to literary or imaginative works that are not based on real events or people. Reality is the world as it is, based on factual information and evidence, while fiction is a product of the imagination that may or may not be inspired by reality.
In many cases, fiction can be a reflection or commentary on reality, and can offer insights into the human experience, societal issues, and cultural values. However, it is important to distinguish between reality and fiction, as the two are fundamentally different and serve different purposes. While reality is concerned with the truth and accuracy of events, fiction is concerned with storytelling, entertainment, and creativity.
It is also worth noting that some works of fiction can blur the line between reality and fiction by incorporating real-world events, people, or places into their narratives, or by presenting fictional scenarios in a way that feels realistic or plausible. However, even in these cases, it is important to maintain a critical awareness of the distinction between reality and fiction." (AI)
"Reality refers to the state of things as they actually exist, while fiction refers to literary or imaginative works that are not based on real events or people. Reality is the world as it is, based on factual information and evidence, while fiction is a product of the imagination that may or may not be inspired by reality.
In many cases, fiction can be a reflection or commentary on reality, and can offer insights into the human experience, societal issues, and cultural values. However, it is important to distinguish between reality and fiction, as the two are fundamentally different and serve different purposes. While reality is concerned with the truth and accuracy of events, fiction is concerned with storytelling, entertainment, and creativity.
It is also worth noting that some works of fiction can blur the line between reality and fiction by incorporating real-world events, people, or places into their narratives, or by presenting fictional scenarios in a way that feels realistic or plausible. However, even in these cases, it is important to maintain a critical awareness of the distinction between reality and fiction." (AI)
102melannen
>100 paradoxosalpha: I suppose that's true, although I will say that as someone who works at (an actually relatively-well funded) public library, those tasks have already been defunded and scraped to the point they mostly consist of getting data from the publishers and entering the data in our databases, which is a job that could probably already be done by some kind of scripting system more efficiently. Our collections staff have admitted they don't really have the time to do anything other than order the books the publishers tell them to order, and nearly all the cataloging is outsourced. They do a little bit of that work manually with things like local interest items that aren't in the larger databases, but chatGPT probably wouldn't be good at that part either.
Deaccessioning is done here mostly by getting the database to spit out a list of books that haven't been checked out in a year or more then weeding on condition, since we don't have the staff for subject matter experts to do it. Our "official job title" librarians mostly do programming, outreach, and management work (as well as reference at the desk!) and are currently super excited about getting chatGPT to give them programming ideas.
Librarians at special-subject archives and corporate libraries and so on probably do more of the cataloging and acquisitions and so on manually, but those kind of places are again probably dealing heavily with the sort of items that chatGPT is unlikely to know how to deal with. Public librarians in very small systems may also do a lot of that manually and personally, but they are also probably working in the sort of jack-of-all trades situation where losing the job to chatGPT is least likely. (It might mean being able to hire someone for the job who doesn't need a graduate degree in cataloging - but public librarianship has been suffering credential inflation for ages away.)
Deaccessioning is done here mostly by getting the database to spit out a list of books that haven't been checked out in a year or more then weeding on condition, since we don't have the staff for subject matter experts to do it. Our "official job title" librarians mostly do programming, outreach, and management work (as well as reference at the desk!) and are currently super excited about getting chatGPT to give them programming ideas.
Librarians at special-subject archives and corporate libraries and so on probably do more of the cataloging and acquisitions and so on manually, but those kind of places are again probably dealing heavily with the sort of items that chatGPT is unlikely to know how to deal with. Public librarians in very small systems may also do a lot of that manually and personally, but they are also probably working in the sort of jack-of-all trades situation where losing the job to chatGPT is least likely. (It might mean being able to hire someone for the job who doesn't need a graduate degree in cataloging - but public librarianship has been suffering credential inflation for ages away.)
