April 2023 - nexa - server-nexa.polito.it

Re: [nexa] Samsung workers made a major error by using ChatGPT. Samsung meeting notes and new source code are now in the wild after being leaked in ChatGPT

by Stefano Zacchiroli

On Thu, Apr 06, 2023 at 10:14:37AM +0200, Marco A. Calamari wrote: > Pur senza avere certezze, non credo che sia possibile effettuare un retraining > incrementale di un LLM. Non conosco il caso specifico di ChatGPT, ma conosco abbastanza bene quello dei large language models (LLMs) per il solo codice sorgente, dei quali ChatGPT è un sovrainsieme. Nello stato dell'arte nessuno fa retraining incrementale di LLM per il codice, anche se è tecnicamente fattibile, perché è troppo costoso in relazione a quanto ci si guadagna in termini di miglioramento dei risultati. Ciò che si fa invece è periodicamente rifare l'addresstamento di tutto il modello da zero, solitamente su dataset che nel frattempo sono molto aumentati di volume (e potenzialmente qualità), e usando pipeline di addestramento che sono anch'esse evolute rispetto alla generazione precedente del LLM. Da un punto di vista di politiche pubbliche, però, trovo che sia assolutamente irrilevante se il (re)training di ChatGPT sia incrementale o rifatto ogni volta a partire da zero. Il punto importante è che OpenAI può tecnicamente stoccare e utilizzare in futuro gli input degli utenti e quindi molto probabilmente lo farà --- nel modo tecnico che trova più adeguato ai bisogni dell'azienda e fino a quando qualcuno, ad esempio un regolatore pubblico, non impedirà loro di farlo. Saluti -- Stefano Zacchiroli . zack(a)upsilon.cc . https://upsilon.cc/zack _. ^ ._ Full professor of Computer Science o o o \/|V|\/ Télécom Paris, Polytechnic Institute of Paris o o o </> <\> Co-founder & CTO Software Heritage o o o o /\|^|/\ https://twitter.com/zacchiro . https://mastodon.xyz/@zacchiro '" V "'

April 6, 2023

Re: [nexa] Samsung workers made a major error by using ChatGPT. Samsung meeting notes and new source code are now in the wild after being leaked in ChatGPT

by Marco A. Calamari

On gio, 2023-04-06 at 09:13 +0200, Stefano Zacchiroli wrote: > Grazie per la segnalazione. Osservo che questo passaggio: > > On Wed, Apr 05, 2023 at 02:09:26PM +0000, Daniela Tafani wrote: > > The upshot is that in just under a month, there were three recorded > > incidences of employees leaking sensitive information via > > ChatGPT. Since ChatGPT retains user input data to further train > > itself, these trade secrets from Samsung are now effectively in the > > hands of OpenAI, the company behind the AI service. > > smentisce alcuni delle ipotesi che sono circolate qui in lista nei > giorni scorsi, sul fatto che ChatGPT *non* usasse gli input per il > (re)training futuro. (Personalmente lo trovo poco sorprendente: non si > capisce perché dovrebbero rinunciare ad utilizzare dati "gentilmente > offerti" dagli utenti del servizio...) Pur senza avere certezze, non credo che sia possibile effettuare un retraining incrementale di un LLM. Penso piuttosto che vengano "riprogrammati" filtraggi" digitali od umani (sfruttati) per modificare/censurare le risposte di GPT-3 in fase di input o dopo l'output. Nessun retraining. GTP continua, come tutti gli LLC a sparare output cinofallici (come direbbe Flora) per sua natura e progetto. JM2EC. Marco > A presto

April 6, 2023

MEMENTO | 161° Mercoledì di Nexa | 12 aprile 2023, ore 17.00

by Nexa Media

Gentilissime, gentilissimi, Vi ricordiamo che *mercoledì 12 aprile*, alle ore 17.00, si terrà il *161° Mercoledì di Nexa* con ospite *Enrico Nardelli* (Università di Roma "Tor Vergata") che presenterà il suo ultimo libro: /*"La rivoluzione informatica. Conoscenza, consapevolezza e potere nella società digitale"*/, Themis, 2022. _L'incontro si terrà IN PRESENZA e ONLINE._ _SEDE FISICA_ dell'incontro: Centro Nexa su Internet e Società, Politecnico di Torino, Via Boggio 65/a, Torino (1° piano). Per accedere alla sala si raccomanda di suonare al citofono portineria e di seguire le indicazioni segnalate dai cartelli apposti all'ingresso e lungo il percorso. QUI <https://nexa.polito.it/contatti> maggiori informazioni su come raggiungerci. _STANZA VIRTUALE_ dell'incontro: https://didattica.polito.it/VClass/NexaEvent Per maggiori informazioni è possibile consultare la pagina: https://nexa.polito.it/mercoledi-161 Cordiali saluti, -- Anita Botta Communication Manager Nexa Center for Internet & Society Politecnico di Torino – DAUIN Corso Duca degli Abruzzi, 24 - 10129 Torino web: https://nexa.polito.it/ mail: anita.botta(a)polito.it tel: 011 090 7219

April 6, 2023

Re: [nexa] Samsung workers made a major error by using ChatGPT. Samsung meeting notes and new source code are now in the wild after being leaked in ChatGPT

by Stefano Zacchiroli

Grazie per la segnalazione. Osservo che questo passaggio: On Wed, Apr 05, 2023 at 02:09:26PM +0000, Daniela Tafani wrote: > The upshot is that in just under a month, there were three recorded > incidences of employees leaking sensitive information via > ChatGPT. Since ChatGPT retains user input data to further train > itself, these trade secrets from Samsung are now effectively in the > hands of OpenAI, the company behind the AI service. smentisce alcuni delle ipotesi che sono circolate qui in lista nei giorni scorsi, sul fatto che ChatGPT *non* usasse gli input per il (re)training futuro. (Personalmente lo trovo poco sorprendente: non si capisce perché dovrebbero rinunciare ad utilizzare dati "gentilmente offerti" dagli utenti del servizio...) A presto -- Stefano Zacchiroli . zack(a)upsilon.cc . https://upsilon.cc/zack _. ^ ._ Full professor of Computer Science o o o \/|V|\/ Télécom Paris, Polytechnic Institute of Paris o o o </> <\> Co-founder & CTO Software Heritage o o o o /\|^|/\ https://twitter.com/zacchiro . https://mastodon.xyz/@zacchiro '" V "'

April 6, 2023

Monopolio della conoscenza [Era: Samsung workers made...]

by Damiano Verzulli

Il 06/04/23 07:00, Guido Vetere ha scritto: > > [...], ci sfugge vero il problema: il costituirsi, sotto i nostri > occhi, del monopolio della conoscenza. mi sfuggono i caratteri di questo "monopolio della conoscenza". Posso chiedere --quando e se possibile-- di dettagliare? Se "osservo" con occhi Italiani o Europei il panorama globale ICT, vedo diverse e numerose forme di "concentrazione" della conoscenza, tutte extra UE. Ma (fortunatamente?) non vedo un unico monopolio (in termini strettamente aziendali). Vedo anche la totale assenza di sforzi nel nostro Paese per prendere coscienza di questo scenario, e provare a mettere in piedi qualche forma di contrasto. È in questo senso che chiedevo di dettagliare. Grazie in anticipo. Un saluto, DV -- Damiano Verzulli e-mail: damiano(a)verzulli.it --- possible?ok:while(!possible){open_mindedness++} --- "...I realized that free software would not generate the kind of income that was needed. Maybe in USA or Europe, you may be able to get a well paying job as a free software developer, but not here [in Africa]..." -- Guido Sohne - 1973-2008 http://ole.kenic.or.ke/pipermail/skunkworks/2008-April/005989.html

April 6, 2023

Ancora su ChatGPT e Garante

by Diego Giorio

Segnalo questo articolo https://www.corrierecomunicazioni.it/privacy/chatgpt-la-provocazione-del-pd… [https://d110erj175o600.cloudfront.net/wp-content/uploads/2023/02/17180239/c…]<https://www.corrierecomunicazioni.it/privacy/chatgpt-la-provocazione-del-pd…> ChatGpt, la "provocazione" del Pd: proposta di legge scritta dall’AI - CorCom<https://www.corrierecomunicazioni.it/privacy/chatgpt-la-provocazione-del-pd…> Atteso in serata l’incontro del Garante Privacy con OpenAi che sostiene di rispettare la privacy e chiede la riattivazione del servizio in Italia. Boom di accesso alle Vpn dopo il blocco: +400%. Intanto l’Autorità britannica multa TikTok per 14,5 milioni: uso illecito dei dati dei minori. L’azienda “esulta” per il dimezzamento della sanzione e assicura: “Under 13 non possono accedere, garantiamo sicurezza” www.corrierecomunicazioni.it Buona giornata e buona Pasqua a tutti D.

April 6, 2023

ChatGPT invented a sexual harassment scandal and named a real law prof as the accused

by Benedetto Ponti

ChatGPT invented a sexual harassment scandal and named a real law prof as the accused The AI chatbot can misrepresent key facts with great flourish, even citing a fake Washington Post article as evidence April 5, 2023 at 2:07 p.m. EDT One night last week, the law professor Jonathan Turley got a troubling email. As part of a research study, a fellow lawyer in California had asked the AI chatbot ChatGPT to generate a list of legal scholars who had sexually harassed someone. Turley’s name was on the list. Tech is not your friend. We are. Sign up for The Tech Friend newsletter. The chatbot, created by OpenAI, said Turley had made sexually suggestive comments and attempted to touch a student while on a class trip to Alaska, citing a March 2018 article in The Washington Post as the source of the information. The problem: No such article existed. There had never been a class trip to Alaska. And Turley said he’d never been accused of harassing a student. A regular commentator in the media, Turley had sometimes asked for corrections in news stories. But this time, there was no journalist or editor to call — and no way to correct the record. (continua qui: https://www.washingtonpost.com/technology/2023/04/05/chatgpt-lies/) --

April 6, 2023

Re: [nexa] Samsung workers made a major error by using ChatGPT. Samsung meeting notes and new source code are now in the wild after being leaked in ChatGPT

by Guido Vetere

La stessa cosa potrebbe accadere postando codice 'sensibile' su piattaforme di supporto alla programmazione come stackoverflow <https://stackoverflow.com/>, con la differenza che da lì si possono rimuovere (non prima ovviamente che il codice in questione sia stato sniffato da qualche migliaio di crawler). Non sappiamo bene come OpenAI usi i dati di conversazione, ma a occhio mi sembra difficile che un utente 'terzo' potrebbe far uscire parti significative di quel codice con qualche prompt. Nello sperimentare ChatGPT come generatore di codice, mi sono accorto che in effetti la sua 'conoscenza' è ferma a un paio di anni fa. Mentre stiamo tutti impegnati nella battaglia sulla riservatezza, da cui OpenAI uscirà indenne, ci sfugge vero il problema: il costituirsi, sotto i nostri occhi, del monopolio della conoscenza. Se ci fosse un Garante del Pluralismo, questo sì che potrebbe far chiudere ChatGPT. G. On Wed, 5 Apr 2023 at 16:09, Daniela Tafani <daniela.tafani(a)unipi.it> wrote: > > Samsung workers have unwittingly leaked top secret data whilst using > ChatGPT to help them with tasks. > > The company allowed engineers at its semiconductor arm to use the AI > writer <https://www.techradar.com/best/ai-writer> to help fix problems > with their source code. But in doing so, the workers inputted confidential > data, such as the source code itself for a new program, internal meeting > notes data relating to their hardware. > > The upshot is that in just under a month, there were three recorded > incidences of employees leaking sensitive information via ChatGPT. Since > ChatGPT retains user input data to further train itself, these trade > secrets from Samsung are now effectively in the hands of OpenAI, the > company behind the AI service. > > In response, Samsung Semiconductor is now developing its own inhouse AI > for internal use by employees, but they can only use prompts that are > limited to 1024 bytes in size. > > In one of the aforementioned cases, an employee asked ChatGPT to optimize > test sequences for identifying faults in chips, which is confidential - > however, making this process as efficient as possible has the potential to > save chip firms considerable time in testing and verifying processors, > leading to reductions in cost too. > > In another case, an employee used ChatGPT to convert meeting notes into a > presentation, the contents of which were obviously not something Samsung > would have liked external third parties to have known. > > Samsung Electronics sent out a warning to its workers on the potential > dangers of leaking confidential information in the wake of the incidences, > saying that such data is impossible to retrieve as it is now stored on the > servers belonging to OpenAI. In the semiconductor industry, where > competition is fierce, any sort of data leak could spell disaster for the > company in question. > > It doesn't seem as if Samsung has any recourse to request the retrieval or > deletion of the sensitive data OpenAI now holds. Some have argued > <https://www.sydney.edu.au/news-opinion/news/2023/02/08/chatgpt-is-a-data-pr…> > that this very fact makes ChatGPT non-compliant with the EU's GDPR, as > this is one of the core tenants of the law governing how companies collect > and use data. It is also one of the reasons why > why Italy has now banned the use of ChatGPT nationwide > <https://www.bbc.co.uk/news/technology-65139406> . > > > > > https://www.techradar.com/news/samsung-workers-leaked-company-secrets-by-us… > > _______________________________________________ > nexa mailing list > nexa(a)server-nexa.polito.it > https://server-nexa.polito.it/cgi-bin/mailman/listinfo/nexa >

April 6, 2023

Victorian mayor readies defamation lawsuit over ChatGPT content

by Daniela Tafani

Victorian mayor readies defamation lawsuit over ChatGPT content Byron Kaye Apr 5, 2023 – 5.37pm <https://subscribe.afr.com/?promote_channel=HI_HL_GNL&utm_source=afr.com&utm…> A Victorian mayor said he may sue OpenAI if it does not correct ChatGPT’s false claims that he had served time in prison for bribery, in what would likely be the first defamation lawsuit against the automated text service. Brian Hood, who was elected mayor of Hepburn Shire, 120 kilometres north-west of Melbourne, last November, became concerned about his reputation when members of the public told him ChatGPT had falsely named him as a guilty party in a foreign bribery scandal involving a subsidiary of the Reserve Bank of Australia<https://www.afr.com/politics/federal/no-rba-cover-up-on-corruption-stevens-…> in the early 2000s. Mr Hood did work for the subsidiary, Note Printing Australia, but was the person who notified authorities about payment of bribes to foreign officials<https://www.afr.com/companies/former-npa-cfo-asks-why-asic-hasn-t-investiga…> to win currency printing contracts, and was never charged with a crime, lawyers representing him said. The lawyers said they sent a letter of concern to ChatGPT owner OpenAI on March 21, which gave OpenAI 28 days to fix the errors about their client or face a possible defamation lawsuit. OpenAI, which is based in San Francisco, had not yet responded to Mr Hood’s legal letter, the lawyers said. OpenAI did not respond to a Reuters email out of business hours. If Mr Hood sues, it would likely be the first time a person has sued the owner of ChatGPT for claims made by the automated language product that has become wildly popular since its launch last year. Microsoft Corp integrated ChatGPT into its search engine Bing in February. A Microsoft spokesperson was not immediately available for comment. “It would potentially be a landmark moment in the sense that it’s applying this defamation law to a new area of artificial intelligence and publication in the IT space,” James Naughton, a partner at Mr Hood’s law firm, Gordon Legal, said. “He’s an elected official, his reputation is central to his role,” Mr Naughton said. Mr Hood relied on a public record of shining a light on corporate misconduct, “so it makes a difference to him if people in his community are accessing this material”. Australian defamation damages payouts are generally capped around $400,000. Mr Hood did not know the exact number of people who had accessed the false information about him – a determinant of the payout size – but the nature of the defamatory statements was serious enough that he may claim more than $200,000, Mr Naughton said. If Mr Hood files a lawsuit, it would accuse ChatGPT of giving users a false sense of accuracy by failing to include footnotes, Mr Naughton said. “It’s very difficult for somebody to look behind that to say ‘how does the algorithm come up with that answer?’” said Mr Naughton. “It’s very opaque.” <https://www.afr.com/technology/victorian-mayor-readies-defamation-lawsuit-o…>

April 5, 2023

Welcome to noyb’s Meta Opt-Out Tool!

by Alberto Cammozzo

<https://meta-out.noyb.eu/> Welcome to noyb’s Meta Opt-Out Tool! The European Data Protection Board (EDPB) has found that Meta has violated the European Data Protection Law (GDPR) by claiming that their personal data collection for advertisement is 'necessary for the performance of a contract' between them and their users. Meta has now announced that it will use another unlawful basis for processing your personal data. Legitimate interests? Instead of giving users a yes / no option, Meta now claims that it has a ‘legitimate interest’ in tracking you, merging your data with the data of others, or analysing your behavior to send you advertisements. This legitimate interest of Meta shall (allegedly) override your right to data protection and privacy – as protected in the Charter of Fundamental Rights of the EU. In simple terms, Meta claims that their interest in making a profit from your personal data overrides your rights. Regulators have previously rejected this approach. Legitimate interests requires option to “opt-out” Under Articles 21(1) and (2) GDPR users can send on “objection” to companies who claim that they have a “legitimate interest” for the processing of their personal data. Basically, this means that users can do an “opt-out”. Meta is providing an opt-out option, but if yo u want to make sure that you send a broad objection that covers the various claims of Meta in one go, you may use this tool to do so. To provide this tool noyb must collect limited information from you, connect with Meta services to verify your account details, and send them to Meta. Meta processes some of your personal data outside of the EU/EEA (including the USA). Read more details here.

April 5, 2023