grazie
ma il punto focale del mio quesito non e' il training ma, prima del training, la genesi
dei testi usati per il training
ciao, s.
On 29/09/23 16:36, Lorenzo Albertini wrote:
§§ 54-64della citazione in giudizio (facilmente reperibile ,_ad es.
qui_<https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=&ved=2ahUKEwiDh_vygtCBAxX4SPEDHZlhAxMQFnoECBYQAQ&url=https%3A%2F%2Fwww.classaction.org%2Fmedia%2Fauthors-guild-et-al-v-openai-inc-et-al.pdf&usg=AOvVaw1tUMb6Gk10kZCsvoAo0PH6&opi=89978449>):
<<54. Recent generative AI systems designed to recognize input text and generate
output text are built on “large language models” or “LLMs.”
55. LLMs use predictive algorithms that are designed to detect statistical
patterns in
the text datasets on which they are “trained” and, on the basis of these
patterns, generate
responses to user prompts. “Training” an LLM refers to the process by which the parameters
that
define an LLM’s behavior are adjusted through the LLM’s ingestion and analysis
of large
“training” datasets.
56. Once “trained,” the LLM analyzes the relationships among words in an input
prompt and generates a response that is an approximation of similar
relationships among words
in the LLM’s “training” data. In this way, LLMs can be capable of generating
sentences,
paragraphs, and even complete texts, from cover letters to novels.
57. “Training” an LLM requires supplying the LLM with large amounts of text for
the LLM to ingest—the more text, the better. That is, in part, the large in large language
model.
58. As the U.S. Patent and Trademark Office has observed, LLM “training” “almost
by definition involve[s] the reproduction of entire works or substantial
portions thereof.”4
59. “Training” in this context is therefore a technical-sounding euphemism for
“copying and ingesting.”
60. The quality of the LLM (that is, its capacity to generate human-seeming
responses
to prompts) is dependent on the quality of the datasets used to “train” the LLM.
61. Professionally authored, edited, and published books—such as those authored
by
Plaintiffs here—are an especially important source of LLM “training” data.
62. As one group of AI researchers (not affiliated with Defendants) has
observed,
“[b]ooks are a rich source of both fine-grained information, how a character, an object or
a scene
looks like, as well as high-level semantics, what someone is thinking, feeling
and how these
states evolve through a story.”5
63. In other words, books are the high-quality materials Defendants want, need,
and
have therefore outright pilfered to develop generative AI products that produce
high-quality
results: text that appears to have been written by a human writer.
64. This use is highly commercial.>>.
_______________
Le informazioni contenute nella presente comunicazione e nei documenti ad essa allegati
potrebbero essere tutelate dal segreto professionale e sono comunque confidenziali e ad
uso esclusivo del destinatario sopra indicato. Qualora la presente comunicazione non fosse
destinata a Voi, Vi preghiamo di tener presente che la divulgazione, distribuzione o
riproduzione di qualunque informazione contenuta nella presente comunicazione o nei
documenti ad essa allegati sono vietate. Se avete ricevuto la presente comunicazione per
errore, Vi preghiamo di volerci avvertire immediatamente e di distruggere quanto ricevuto
senza leggerlo. Grazie per la collaborazione.
The information contained in this email and any documents attached to it may be legally
privileged and confidential. The information is intended only for the use of the
individual or entity named above. If you are not the intended recipient, you are hereby
notified that any use, dissemination, distribution or reproduction of any information
contained in or attached to this email is prohibited. If you have received this email in
error, please immediately notify us by reply email or by telephone, and destroy the
original transmission and its attachments without reading them. Thank you.
-----Messaggio originale-----
Da: nexa <[email protected]> Per conto di Rossana
Morriello
Inviato: venerdì 29 settembre 2023 16:08
A: Nexa <[email protected]>
Oggetto: [nexa] R: ‘Biggest act of copyright theft in history’: thousands of
Australian books allegedly used to train AI model | Australia news | The
Guardian
Non sono una giurista ma credo che questa rassegna possa essere utile alla
discussione
https://www.thefashionlaw.com/from-chatgpt-to-deepfake-creating-apps-a-<https://www.thefashionlaw.com/from-chatgpt-to-deepfake-creating-apps-a-running-list-of-key-ai-lawsuits/>
running-list-of-key-ai-lawsuits/
Saluti
Rossana Morriello
-----Messaggio originale-----
Da: nexa <[email protected]<mailto:[email protected]>>
Per conto di Stefano
Quintarelli
Inviato: venerdì 29 settembre 2023 15:21
Cc: Nexa <[email protected]<mailto:[email protected]>>
Oggetto: Re: [nexa] ‘Biggest act of copyright theft in history’: thousands of
Australian books allegedly used to train AI model | Australia news | The
Guardian
Ho una domanda per i giuristi (anzi, piu' di una)
per allenare un modello, ho bisogno di un file con la versione digitale di un
testo.
(cosnsidero ovviamente testi non PD, CC0, ecc.)
la versione digitale di un testo la posso ottenere da un ebook (gia' digitale),
togliendo il probabile DRM.
ma un ebook non e' unbene ma e' un servizio soggetto a licenza d'uso, quindi
se non e'
prevista nella licenza d'uso la facolta' di estrarre il testo digitale per
allenarci un
modello, mi sembra che ci sia gia' una violazione della licenza, per cui, credo,
non possa essere usato come base di un allenamento, tanto piu' se il fine di
tale allenamento e'
commerciale (se vendo un servizio basato su quel modello).
se e' cosi', per allenare il mio modello devo allora prednere il testo digitale
facendo scan/ocr di un testo cartaceo.
ma cio' e' possibile, se non erro, solo per uso personale e non commerciale.
se questo e' corretto, non mi pare ci sia un modo per prendere un testo digitale
senza infrangere una licenza d'uso/copyright
dove e' la fallacia del ragionamento ?
grazie, s.
On 29/09/23 15:00, Stefano Borroni Barale wrote:
> Buongiorno lista,
>
>> L'idea che istruire un modello su dei testi coperti da copyright sia
>> una violazione del suddetto copyright è altamente opinabile
>
> Fin qui, ho l'impressione che tutti i legali in lista concorderanno.
>
>> ragionamento è in realtà abbastanza semplice: se istruirsi su un
>> testo ne violasse il copyright, saremmo tutti dei criminali.
>
> Ma siccome noi siamo umani e quello che produciamo non è - salvo i discorsi
dei politici(*) - ontologicamente identico alla produzione di esseri tecnici non
viventi, logica vuole che quanto si applica a noi non possa applicarsi a un LLM,
tanto quanto la legge sul copyright non si applica pedissequamente all'utilizzo
di testi umani per creare modelli linguistici.
>
> Questo è il motivo per il quale tutti i tentativi di "proteggere via
copyright" il
prodotto di software generativi sono falliti miseramente, e con motivazioni
scritte in sentenze; che per il diritto credo abbiano un peso assai maggiore del
sito di CC.
>
> La mia impressione è che la questione terrà impegnati legali, informatici,
filosofi e società ancora moooooolto a lungo.
> SBB
>
> (*) Come sanno bene i bambini degli anni '80 che hanno giocato con
> questo spassoso giocattolo:
>https://www.enricodalbosco.it/giochi/tubolario/<https://www.enricodalbosco.it/giochi/tubolario/>
>
>
> Di quei testi
>> non c'è fisicamente traccia all'interno dei modelli, non viene
>> copiato niente. I modelli sono un'opera trasformativa di quei testi,
>> non derivativa.
>>
>> Lo argomenta molto bene Creative Commons:
>>https://creativecommons.org/2023/02/17/fair-use-training-generative-a<https://creativecommons.org/2023/02/17/fair-use-training-generative-a>
>> i/
>>
>> Detto questo, cito le parole di un altro autore, Jeff Jarvis:
>>
https://www.facebook.com/jeff.jarvis/posts/pfbid0LMFeqdTYoxnGHQAZwp5<https://www.facebook.com/jeff.jarvis/posts/pfbid0LMFeqdTYoxnGHQAZwp5H>
H
>> MmeeVqgMSjL2dkcwMcBojkb2cinBpgYTHyc7Fhq1B9NPl
>>
>> «I, for one, am not complaining about my books being in in large
>> language model training sets. I write to enter ideas into public
>> discourse. I prefer informed over ignorant AI. I believe it is fair
>> use for anyone to read & use books for transformative work. In fact,
>> I'd probably feel snubbed if my books were not there. I'm happy when
>> they are in libraries. I'm fine that they're here.»
>>
>> Fabio
>>
>> Il giorno ven 29 set 2023 alle ore 07:52 Alberto Cammozzo via nexa
>>[email protected]<mailto:[email protected]>ha scritto:
>>
>>>https://www.theguardian.com/australia-news/2023/sep/28/australian-<https://www.theguardian.com/australia-news/2023/sep/28/australian-bo>
bo
>>> oks-training-ai-books3-stolen-pirated
>>>
>>> Thousands of books from some of Australia’s most celebrated authors
have potentially been caught up in what Booker prize-winning novelist Richard
Flanagan has called “the biggest act of copyright theft in history”.
>>>
>>> The works have allegedly been pirated by the US-based Books3 dataset
and used to train generative AI for corporations such as Meta and Bloomberg.
>>>
>>> Flanagan, who found 10 of his works, including the multi-international
award-winning 2013 novel The Narrow Road to the Deep North, on the
Books3 dataset, told Guardian Australia he was deeply shocked by the
discovery made several days ago.
>>>
>>> “I felt as if my soul had been strip mined and I was powerless to stop it,”
he said in a statement.
>>>
>>> “This is the biggest act of copyright theft in history.”
>>>
>>> AI could ‘turbo-charge fraud’ and be monopolised by tech companies,
>>> Andrew Leigh warns
>>>
>>> The Australian Publishers Association confirmed to Guardian Australia on
Wednesday that as many as 18,000 fiction and nonfiction titles with
Australian ISBNs (unique international standard book numbers) appeared to
be affected by the copyright infringement, although it is not yet clear what
proportion of these are Australian editions of internationally authored books.
>>>
>>> “We’re still working through [the data] to work out the impact in terms of
Australian authors,” APA spokesperson Stuart Glover said.
>>>
>>> “This is a massive legal and ethical challenge for the publishing industry
and for authors globally.”
>>>
>>> A search tool published on Monday by US media platform The Atlantic and
uploaded by the US Authors Guild on Wednesday revealed the works of Peter
Carey, Helen Garner, Kate Grenville, Anna Funder, Christos Tsiolkas and
Thomas Keneally, as well as Flanagan and dozens of other high-profile
Australian authors, were included in the pirated dataset containing more than
180,000 titles.
>>>
>>> On Thursday, the Australian Society of Authors issued a statement saying
it was “horrified” to learn that the works of Australian writers were being used
to train artificial intelligence without permission from the authors.
>>>
>>> ASA chief executive, Olivia Lanchester, described the Books3 dataset as
piracy on an industrial scale.
>>>
>>> “Authors appropriately feel outraged,” Lanchester said. “The fact is this
technology relies upon books, journals, essays written by authors, yet
permission was not sought nor compensation granted.”
>>>
>>> Lanchester said the Australian literary industry, while not objecting per se
to emerging technologies such as AI, was deeply concerned about the lack of
transparency evident in the development and monetisation of AI by global
tech companies.
>>>
>>> “Turning a blind eye to the legitimate rights of copyright owners threatens
to diminish already precarious creative careers,” she said.
>>>
>>> “The enrichment of a few powerful companies is at the cost of thousands
of individual creators. This is not how a fair market functions.”
>>>
>>> Josephine Johnston, chief executive of Australia’s Copyright Agency,
described the Books3 development as “a free kick to big tech” at the expense
of Australia’s creative and cultural life.
>>>
>>> “We’re going to need greater transparency – how these tools have been
developed, trained, how they operate – before people can truly understand
what their legal rights might be,” she said.
>>>
>>> “We seem to be in this terrible position now where content owners –
remembering that the vast majority of them will be individual authors – may
actually have to take out court cases to enforce their rights.”
>>>
>>> Australian copyright law protects creators of original content from data
scraping.
>>>
>>> Litigation in the US against ChatGPT creator OpenAI over use of allegedly
pirated book datasets, Books1 and Books2 (which do not appear to be
affiliated with Books3) has already commenced.
>>>
>>> In July, North American horror/fantasy writers Mona Awad (author of
Bunny) and Paul Tremblay (author of The Cabin at the End of the World) filed a
lawsuit in a San Francisco federal court, alleging ChatGPT unlawfully digested
their books as part of its AI training data.
>>>
>>> On 28 August, OpenAI filed a motion to dismiss the lawsuit, arguing that
the authors “misconceive the scope of copyright, failing to take into account
the limitations and exceptions (including fair use) that properly leave room for
innovations like the large language models now at the forefront of artificial
intelligence”.
>>>
>>> On 19 September the Writers Guild and 17 of its members, including
bestselling novelists John Grisham, George RR Martin and Jodi Picoult, filed a
complaint in a New York district court against OpenAI, seeking redress for
“flagrant and harmful infringements” of guild members’ registered copyrights.
>>>
>>> In a statement on its website, the guild says while it is aware that
companies such as Meta and Bloomberg have used the Books3 dataset to
train their LLMs, it is not yet clear whether OpenAI is using Books3 to train
its
ChatGPT models GPT 3.5 or GPT 4.
>>>
>>> Democracies face ‘truth decay’ as AI blurs fact and fiction, warns
>>> head of Australia’s military
>>>
>>> Guardian Australia has sought comment from OpenAI, which has yet to
officially respond to the guild’s complaint, and Meta.
>>>
>>> On 4 September, US technology magazine Wired reported that a Danish
anti-piracy group called Rights Alliance had been told by Bloomberg that the
company did not plan to train future versions of its BloombergGPT using
Books3.
>>>
>>> Bloomberg declined to respond to the Guardian’s queries.
>>>
>>> The APA said the global nature of the issue would present significant
challenges in enforcement and prosecution, and has joined the authors’
society in calling for AI technologies to be regulated.
>>>
>>> Consultation closed last month for a Department of Industry, Science and
Resources discussion paper on supporting responsible AI.
>>>
>>> A parliamentary inquiry is under way examining the use of generative
artificial intelligence in the Australian education system.
>>>
>>> Flanagan said it was up to the Australian government to act to protect
Australia’s writers.
>>>
>>> “It has power and we do not,” he said.
>>>
>>> “If it cares for our culture it must now stand up and fight for it.”
>>>
>>> _______________________________________________
>>> nexa mailing list
>>>[email protected]<mailto:[email protected]>
>>>https://server-nexa.polito.it/cgi-bin/mailman/listinfo/nexa<https://server-nexa.polito.it/cgi-bin/mailman/listinfo/nexa>
>>
>> _______________________________________________
>> nexa mailing list
>>[email protected]<mailto:[email protected]>
>>https://server-nexa.polito.it/cgi-bin/mailman/listinfo/nexa<https://server-nexa.polito.it/cgi-bin/mailman/listinfo/nexa>
> _______________________________________________
> nexa mailing list
>[email protected]<mailto:[email protected]>
>https://server-nexa.polito.it/cgi-bin/mailman/listinfo/nexa<https://server-nexa.polito.it/cgi-bin/mailman/listinfo/nexa>
_______________________________________________
nexa mailing list
[email protected]<mailto:[email protected]>
https://server-nexa.polito.it/cgi-bin/mailman/listinfo/nexa<https://server-nexa.polito.it/cgi-bin/mailman/listinfo/nexa>
_______________________________________________
nexa mailing list
[email protected]<mailto:[email protected]>
https://server-nexa.polito.it/cgi-bin/mailman/listinfo/nexa<https://server-nexa.polito.it/cgi-bin/mailman/listinfo/nexa>
_______________________________________________
nexa mailing list
[email protected]
https://server-nexa.polito.it/cgi-bin/mailman/listinfo/nexa
_______________________________________________
nexa mailing list
[email protected]
https://server-nexa.polito.it/cgi-bin/mailman/listinfo/nexa
