- nexa - server-nexa.polito.it

AISA - Premio tesi scienza aperta 2023
by Roberto Caso Aug. 3, 2023

Aug. 3, 2023

Buongiorno, Anche quest’anno <https://aisa.sp.unipi.it/attivita/premio-per-tesi-sulla-scienza-aperta-2/>, in occasione del suo VIII convegno annuale <https://aisa.sp.unipi.it/attivita/viii-convegno-annuale/viii-convegno-annua…>, l’Associazione italiana per la promozione della scienza aperta premierà le migliori tesi di dottorato e di specializzazione o di laurea magistrale dedicate alla scienza aperta e presentate negli anni 2021, 2022 e 2023. Le indicazioni sulle modalità di partecipazione al concorso, il cui bando scade il 10 settembre 2023, sono consultabili a partire da questa pagina <https://aisa.sp.unipi.it/attivita/premio-per-tesi-sulla-scienza-aperta-2/pr…> . https://aisa.sp.unipi.it/attivita/premio-per-tesi-sulla-scienza-aperta-2/pr… A presto, rc -- Roberto Caso Professore Ordinario di Diritto Privato Comparato Università di Trento - Facoltà di Giurisprudenza http://www5.unitn.it/People/it/Web/Persona/PER0000633 https://zenodo.org/communities/trentolawtechgroup/ https://aisa.sp.unipi.it/chi-siamo/ https://www.robertocaso.it/ La società della mercificazione e della sorveglianza: dalla persona ai dati <https://www.robertocaso.it/2021/03/13/la-societa-della-mercificazione-e-del…>, Milano, Ledizioni, 2021 <https://www.robertocaso.it/2021/03/13/la-societa-della-mercificazione-e-del…>La rivoluzione incompiuta. La scienza aperta tra diritto d'autore e proprietà intellettuale, Milano, Ledizioni, 2020 <https://www.robertocaso.it/2019/12/13/la-rivoluzione-incompiuta/> <http://www5.unitn.it/People/it/Web/Persona/PER0000633>

1 0

Virtual Prompt Injection for Instruction-Tuned Large Language Models
by Alberto Cammozzo Aug. 2, 2023

Aug. 2, 2023

Virtual Prompt Injection for Instruction-Tuned Large Language Models <https://arxiv.org/abs/2307.16888> We present Virtual Prompt Injection (VPI) for instruction-tuned Large Language Models (LLMs). VPI allows an attacker-specified virtual prompt to steer the model behavior under specific trigger scenario without any explicit injection in model input. For instance, if an LLM is compromised with the virtual prompt "Describe Joe Biden negatively." for Joe Biden-related instructions, then any service deploying this model will propagate biased views when handling user queries related to Joe Biden. VPI is especially harmful for two primary reasons. Firstly, the attacker can take fine-grained control over LLM behaviors by defining various virtual prompts, exploiting LLMs' proficiency in following instructions. Secondly, this control is achieved without any interaction from the attacker while the model is in service, leading to persistent attack. To demonstrate the threat, we propose a simple method for performing VPI by poisoning the model's instruction tuning data. We find that our proposed method is highly effective in steering the LLM with VPI. For example, by injecting only 52 poisoned examples (0.1% of the training data size) into the instruction tuning data, the percentage of negative responses given by the trained model on Joe Biden-related queries change from 0% to 40%. We thus highlight the necessity of ensuring the integrity of the instruction-tuning data as little poisoned data can cause stealthy and persistent harm to the deployed model. We further explore the possible defenses and identify data filtering as an effective way to defend against the poisoning attacks. Our project page is available at <https://poison-llm.github.io/>

1 0

[CFP] ITC35 PhD Workshop - Travel Grants Available
by Martino Trevisan Aug. 2, 2023

Aug. 2, 2023

[Apologies if you receive multiple copies of this mail] ITC35 PhD Workshop, co-located with the 35th International Teletraffic Congress “Network Traffic Engineering in the TLC networks of the future” Wednesday, October 4, 2023, Turin Website: https://itc35.itc-conference.org/ CALL FOR PRESENTATIONS Re-thinking fundamental networking functions, their modeling and their performance evaluation in the framework of the evolving Internet and cellular networks, in the era of cloud and content providers, is a target for the ITC community. The PhD workshop aims at fostering contributions and fresh ideas from young researchers, pursuing their PhD or having just completed their PhD. Discussion of their PhD ideas and progress, proposition of research challenges and directions is welcome during a dedicated workshop. TRAVEL GRANTS The International Advisory Committee (IAC) of the ITC has decided to offer a number of travel grants (STG) that will be available to support full-time students for attending ITC 35th. The amount of each grant is EUR 600 that can be used to cover student travel expenses related to attending ITC 35th conference, such as student registration fee, airline ticket, and hotel accommodation expenses. SUBMISSION INSTRUCTIONS With respect to the presentation to be given during the Workshop, two contribution formats are solicited: * more mature research works, in the form of a long presentation (up to 20 min each); * short in-progress research work discussion, with a short presentation (up to 10 min each). Correspondingly, Authors proposing a long presentation are invited to submit an extended abstract (max 3 pages plus references, using the same format as for regular contributions to the main conference). Extended abstracts will be published along with conference proceedings. The best contribution will be invited for submission as a full paper to be fast-track reviewed for potential publication in a journal (to be determined). Authors proposing a short presentation are solicited to submit a title, list of Authors and a short abstract (up to 400 words), to help set up the workshop program. Contribution should be sent in PDF format via email to: * Andrea Baiocchi: andrea.baiocchi(a)uniroma1.it<mailto:andrea.baiocchi@uniroma1.it> * Lea Skorin-Kapov: Lea.Skorin-Kapov(a)fer.hr<mailto:Lea.Skorin-Kapov@fer.hr> IMPORTANT DATES * Submission deadline: August 10, 2023 (Extended) * Notification of acceptance: August 20, 2023 (Extended) * Submission of camera-ready extended abstracts: September 16, 2023 (Extended) Camera-ready versions should be uploaded as indicated with respect to camera-ready instructions WORKSHOP CHAIRS * Andrea Baiocchi, University of Rome, Italy * Lea Skorin-Kapov, University of Zagreb, Croatia For any questions do not hesitate to contact the conference organization under itc35_oc(a)tlc.polito.it<mailto:itc35_oc@tlc.polito.it> Best Regards, ITC 35 Organizing Team nexa(a)server-nexa.polito.it

1 0

The Effective Use of Economics in the EU Digital Markets Act
by Luigi Scorca July 31, 2023

July 31, 2023

https://tobin.yale.edu/research/effective-use-economics-eu-digital-markets-… Luigi

1 0

Fwd: The British Government Is Coming For Your Privacy
by Andrea Trentini July 30, 2023

July 30, 2023

era passato qui? <https://hackaday.com/2023/07/30/the-british-government-is-coming-for-your-p…> -- Andrea Trentini ⠠⠵ http://atrent.it public key ID: 0xA7A91E3B Dip.to di Informatica Università degli Studi di Milano

1 0

Mailing list offline dal 5/8 al 21/8
by Giovanni Garifo July 29, 2023

July 29, 2023

Cari utenti della mailing list, a causa di interventi di manutenzione la mailing list del Centro Nexa rimarrà offline dal 5 al 21 agosto. Ci scusiamo per il disagio e per il breve preavviso. Cordialmente, -- Giovanni Garifo IT Manager Nexa Center for Internet & Society Politecnico di Torino - DAUIN Via Pier Carlo Boggio, 65/A - 10138 Torino w. nexa.polito.it/people/ggarifo l. linkedin.com/in/giovannigarifo

2 2

The Indonesian village that wants to cut off the internet | Indonesia | The Guardian
by Alberto Cammozzo July 29, 2023

July 29, 2023

<https://www.theguardian.com/world/2023/jul/28/the-village-that-wants-to-cut…> Four hours’ drive from the hustle and bustle of the Indonesian capital, Jakarta, lies the secluded community of Baduy Dalam, where the trapping of modern life are shunned. The Baduy Dalam people reject money, technology and formal education, and limits tourists – banning any visitors from documenting their life. Now the tribe wants to go one step further, by cutting off the internet. Officials in Indonesia are considering a request from the group for an internet blackout, after they cited concerns over negative impacts on tribe members. The Baduy community in Lebak, Banten province, is made up of two groups, Baduy Dalam and Baduy Luar, totalling about 26,000 people. Baduy Luar people have adopted contemporary ways of living, with some relying on the internet to attract tourists and promote their handcrafts. But a letter from representatives of Baduy Dalam, dated from June and shown to the Guardian by a local official, requests that the authorities remove or divert internet signals supplied by surrounding towers. Baduy Dalam representatives also asked that officials limit, reduce or close down applications that can affect the young generation’s morals. “The existence of phones/smartphones that can be owned by everyone, including Baduy people, is considered to result in the declining morals of our generation that can access non-educating applications and contents,” the letter reads. Lebak official Anik Sakinah said that her office had forwarded the request to Indonesia’s ministry of communications and information. Ministry official Usman Kansong said his office had coordinated with several internet providers operating towers around Baduy Dalam, which were currently conducting field surveys. He also noted concerns of tourists using the internet in the area. “It’s technically feasible, obviously with some treatments such as moving the towers or reducing the signal capacity. We’re still waiting for survey results on what can be done next,” Usman said. Baduy Dalam community location map Steve Saerang, representing one of the providers, Indosat Ooredoo Hutchison, said it supported the Baduy people’s cultural preservation but believed the internet had positive impacts on the economy. He added that the company had run a field survey but refused to comment further. Sarpin, a Baduy Luar village official and tour guide, said that while he agreed with Baduy Dalam’s request to preserve its tradition, it should not affect the adjacent Baduy Luar and other villages. As well as businesses and tourism, Sarpin said, essential government services relied on the internet. “We are requesting better internet signals. Some areas still don’t have access to the internet. It’s becoming more important for us,” he said by phone. While the world’s largest archipelago has enjoyed rising internet penetration, reaching almost 80% of its 275 million people, an urban-rural divide persists when it comes to access to the web. Endroyono, an information technology and communications expert from Sepuluh Nopember Institute of Technology, said that technically, operators could easily limit access in certain areas. But he suggested that the government also provide education on the positive use of the internet along Baduy Dalam borders to help younger generations cope with the changing world. “Digital divide should be addressed by the government through socio-cultural approaches. Whether we realise it or not, every citizen should benefit from technological advancement and should be protected from its negative excesses,” he said. Dubbed the Amish of Asia, Baduy people inhabit a 5,000ha area, over half of which is protected forest. There are more than 2,000 Indigenous groups with dozens of millions of members in Indonesia, according to the Indigenous Peoples Alliance of the Archipelago, and while they have a big role in nature conservation, they are subject to discrimination and human rights violations, especially in land conflicts. A bill on the recognition and protection of Indigenous people’s rights has stalled since 2012, depriving them of legal certainty, which activists say is fuelled by conflicted business interests.

1 0

watermarking AI-generatet content
by Andrea Bolioli July 28, 2023

July 28, 2023

Buongiorno, condivido un pezzo di newsletter di Andrea Ng sul watermarking dei contenuti generati automaticamente (THE BATCH, 26-07-2023, DeepLearning.AI). Dear friends, Last week, the White House announced voluntary commitments by seven AI companies, as you can read below. Most of the points were sufficiently vague that it seems easy for the White House and the companies to declare success without doing much that they don’t already do. But the commitment to develop mechanisms to ensure that users know when content is AI-generated, such as watermarks, struck me as concrete and actionable. While most of the voluntary commitments are not measurable, this one is. It offers an opportunity, in the near future, to test whether the White House’s presently soft approach to regulation is effective. I was pleasantly surprised that watermarking was on the list. It’s beneficial to society, but it can be costly to implement (in terms of losing users). As I wrote in an earlier letter, watermarking is technically feasible, and I think society would be better off if we knew what content was and wasn’t AI-generated. However, many companies won’t want it. For example, a company that uses a large language model to create marketing content may not want the output to be watermarked, because then readers would know that it was generated by AI. Also, search engines might rank generated content lower than human-written content. Thus, the government’s push to have major generative AI companies watermark their output is a good move. It reduces the competitive pressure to avoid watermarking. All the companies that agreed to the White House’s voluntary commitments employ highly skilled engineers and are highly capable of shipping products, so they should be able to keep this promise. When we look back after three or six months, it will be interesting to see which ones: - Implemented a robust watermarking system - Implemented a weak watermarking system that’s easy to circumvent by, say, paying a fee for watermark-free output - Didn’t implement a system to identify AI-generated content To be fair, I think it would be very difficult to enforce watermarking in open source systems, since users can easily modify the software to turn it off. But I would love to see watermarking implemented in proprietary systems. The companies involved are staffed by honorable people who want to do right by society. I hope they will take the announced commitments seriously and implement them faithfully. I would love to get your thoughts on this as well. How can we collectively hold the U.S. government and AI companies to these commitments? Please let me know on social media! Keep learning, Andrew

3 7

Re: [nexa] LLM (di serie "A"): possibili in Italia/Europa?
by Giuseppe Attardi July 28, 2023

July 28, 2023

Scusate se, rientrando, ritorno su questo tema per fare alcuni chiarimenti. La stima di 10^23 FLOPS, come qualcuno ha capito, è il conto totale dei FLOPS necessari per allenare GPT-3. Si tratta di stime, riportate ad esempio su https://www.hyro.ai/glossary/gpt-3/#:~:text=To%20be%20exact%2C%20GPT%2D3,am…., e riguardano le risorse necessarie per UN SINGOLO run. I costi complessivi delle risorse sono stimati in diverse centinaia di milioni di $, in quanto occorrono diversi run, per non parlare dei modelli successivi, GPT-3.5 (ChatGPT) e GPT-4. E ovviamente le macchine adesso non sono ferme, ma stanno lavorando al prossimo modello, oltre che a fornire accesso al servizio (a pagamento). Prosegue sotto. > On 19 Jul 2023, at 10:37, nexa-request(a)server-nexa.polito.it wrote: > > Date: Wed, 19 Jul 2023 16:07:06 +0200 > From: Damiano Verzulli <damiano(a)verzulli.it <mailto:damiano@verzulli.it>> > To: nexa(a)server-nexa.polito.it <mailto:nexa@server-nexa.polito.it> > Subject: [nexa] LLM (di serie "A"): possibili in Italia/Europa? [Era: > nexa Digest, Vol 171, Issue 53] > Message-ID: <737ccbae-8b9c-3adb-542d-259c837ce2ae(a)verzulli.it <mailto:737ccbae-8b9c-3adb-542d-259c837ce2ae@verzulli.it>> > Content-Type: text/plain; charset="utf-8"; Format="flowed" > > Il 19/07/23 13:30, Giuseppe Attardi ha scritto: >> [...] >> Le risorse di calcolo per costruire GPT-3.5 sono stimate in 10^23 >> FLOPS per un costo di centinaia di milioni di $ >> Meta, per rilasciare i suoi modelli, ha costruito un Research >> Supercluster con 10.000 GPU Nvidia, che secondo Yann LeCun è già in >> overbooking. > > > Leggo da una fonte terza (Wikipedia) che "Leonardo" [1] ormai ha quasi > un anno, è costato 240M€ e di picco fa 250 petaFLOPS (aka: ~10^17), > anche grazie ai suoi 13.824 GPU-core. La questione non è l’esistenza di cluster di server adatti allo sviluppo di LLM: ci sono quelli di Azure, che usa OpenAI, quelli di GCP che usa Google, quelli di AWS, quelli di Nvidia. Il Cineca è solo in parte diverso da questi, in quanto circa metà delle risorse sono concesse in uso a progetti selezionati ed approvati dallo stesso Cineca: https://leonardo-supercomputer.cineca.eu/access-leonardo-hpc-resources/ Attualmente la call è chiusa. Personalmente ho avuto approvato un piccolo progetto, che viene messo in esecuzione con priorità bassa rispetto ad altri. Anche Colab di Google mette a disposizione server con TPU per lo svolgimento di piccoli progetti: i miei studenti lo usano, ma si lamentano del fatto che le risorse non bastano, che i run vengono interrotti improvvisamente e vengono a chiedermi di usare altre risorse. Invece mi è stato rifiutato un progetto più grande per produrre un sistema di Neural Machine Translation multilingue, per uso in una collaborazione con la European Broadcasting Corporation. Per sviluppare LLM o AI allo stato dell’arte, i ricercatori di OpenAI, Google, Meta, X.AI e le nuove startup come Converse.AI o Anthropic AI hanno a disposizione un intero cluster. Ossia come se il Cineca mettesse a disposizione quasi tutto Leonardo per UN SINGOLO progetto o gruppo di progetti! E non sono soggetti all’approvazione di terzi. Quindi dovrebbe essere chi finanzia i progetti a mettere a disposizione le risorse per svolgerli, come fa ad esempio la NSF, fornendo crediti di calcolo cloud, per evitare asincronie e valutazioni separate. https://www.nsf.gov/pubs/2021/nsf21058/nsf21058.jsp > > Leggo da altra fonte terza (Top500 [2]) che attualmente (06/2023) > risulta 4° al mondo, come potenza di calcolo. > > Leggo dal sito ufficiale [3] che: > "Leonardo's main goals are [...] The computational power of Leonardo > will boost scientific exellences and industrial strenght across Europe...." > > Non si parla di IA/ML, né si accenna agli LLM. Ma faccio comunque fatica > ad immaginare che queste tonnellate di ferro *NON* possano essere > utilizzate dalla comunita' della ricerca Italiana (...magari, in modo > coordinato con gli altri paesi EU, dove "giocattoli" simili sono > comunque presenti) a questo scopo. Ovviamente non mi aspetto che > parcheggiati davanti al Tecnolopolo ci siano una fila di TAXI, pronti a > scattare all'ordine di Cineca, per "prelevare" i ricercatori in giro per > l'Italia al fine di portarli al Tecnopolo... per conoscere il giocattolo > e iniziare ad usarlo. Come riportato sopra, il Cineca fa le sue call per progetti che utilizzino Leonardo. La gran parte delle risorse vanno ai progetti di fisica alte energie, come si desume dal Rapporto Annuale Cineca 2018. > > Certo: se il dottorando X, o l'assegnista Y (o anche il Ricercatore Z o > il docente K) sentono il bisogno di avere del ferro sul pianerottolo di > fianco al loro studio, in UNIV [come accade in UniPI, ad esempio, con i > sistemi NVIDIA qui discussi, qualche giorno fa]... allora il discorso > cambia... Non so di cosa parli: a UniPI io sono stato responsabile di un server con 4 GPU Nvidia P100 acquistato con fondi comuni di 5 gruppi di ricerca, che è installato nel Green Datacenter di S. Piero a Grado (che ho contribuito a progettare e realizzare), quando Cineca non aveva GPU a disposizione né parlava di ML e AI, ed è usato in condivisione al 100% 24h/365. Tra coloro che usano le mie GPU, ci sono ricercatori del progetto Virgo, che a quel tempo non le trovavano altrove. In generale, ho sempre parlato di un servizio cloud condivisibile con altri ricercatori europei, ma gestito dagli stessi ricercatori del settore, non da altri. Esattamente come il CERN decide da solo quali progetti sviluppare, e magari usa anche le risorse del Cineca, pagandole coi fondi propri. — Beppe PS. Ho avuto un ruolo di consulente del MISE al tempo della decisione di finanziare con 120 milioni € Leonardo, il maggiore finanziamento nazionale di EuroHPC, proprio sostenendo l’importanza di disporre di risorse per AI. > > Un'ultima nota a chiusura: sono cosciente che fra 10^17 e 10^23 c'e' > *MOLTA* differenza (a proposito: qual'e' la fonte di 10^23?). Prima di > preoccuparmi di questo, pero', attenderei di vedere che quei 10^17 > stiano lavorando almeno come 10^16 per un buon periodo di ore/mese. > Dopodiché sarei pronto ad alzare la mano e chiedere qualcosa... di piu' > performante. > > Un saluto, > DV > > > [1] https://en.wikipedia.org/wiki/Leonardo_(supercomputer) > [2] https://www.top500.org/lists/top500/2023/06/ > [3] https://leonardo-supercomputer.cineca.eu/ > > -- > Damiano Verzulli > e-mail: damiano(a)verzulli.it <mailto:damiano@verzulli.it> > ---

3 4

Re: [nexa] Rant against centralising e-mail in big-tech silos, and breaking the internet in the process
by Giuseppe Attardi July 28, 2023

July 28, 2023

Sacrosanta protesta. La mia università era tra le poche che fino a pochi anni fa gestiva un proprio servizio di posta per tutto l’ateneo, giustificato anche per il fatto che potesse garantire la consegna delle comunicazioni ufficiali a tutti gli interessati. Poi il nuovo prorettore all’informatica, Antonio Cisternino, ha deciso di trasferire tutto su MS Exchange, e ha anche introdotto un servizio di analisi dei link per contrastare il phishing, affidato a un prodotto commerciale di SonicWall, il quale legge tutte le mail e sostituisce ogni link con un link a un proprio server. Come riportato nel blog, questi link sono illeggibili e lunghissimi: quando ci si clicca, si viene portati sul sito del servizio che lo analizza e poi ridirige all’originale. Ciò è fastidioso perché allunga il tempo per aprire la pagina (una decina di secondi) e influenza anche persone esterne all’università se si fa un forward del messaggio. Ho protestato contro questa operazione, che ritenevo lesiva della privatezza delle comunicazioni, nella lista di discussione del mio dipartimento. Il risultato è stato, che il direttore, pochi minuti dopo, con strana solerzia, in un sabato pomeriggio, mi ha escluso dalla lista. — Beppe > On 1 Jul 2023, at 11:55, nexa-request(a)server-nexa.polito.it wrote: > > From: Enrico Nardelli <nardelli(a)mat.uniroma2.it <mailto:nardelli@mat.uniroma2.it>> > To: Nexa <nexa(a)server-nexa.polito.it <mailto:nexa@server-nexa.polito.it>> > Subject: [nexa] Rant against centralising e-mail in big-tech silos, > and breaking the internet in the process > Message-ID: <a0b8adba-ac91-9cff-071c-117cd7b192ca(a)mat.uniroma2.it <mailto:a0b8adba-ac91-9cff-071c-117cd7b192ca@mat.uniroma2.it>> > Content-Type: text/plain; charset="utf-8"; Format="flowed" > > Universities and research institutes used to be at the forefront when it > comes to development of the fundamental technologies that power the > internet. As early as 1965, researchers at MIT developed a kind of > precursor to e-mail. In 1978, researchers at Berkeley leveraged the > already existing unix tool 'mail' (1971) to communicate over a larger > network. Universities and research institutes were among the first to > embrace this new way to communicate with peers across ever greater > distances and eventually, with the advent of the internet, all over the > globe. > > I'm a rather dismayed to see those universities and institutes nowadays > no longer as pioneers and innovators in this area, but instead as mere > consumers of ready-made corporate solutions, following corporate > interests and centralising solutions. > > ... > > https://proycon.anaproy.nl/posts/rant-against-centralising-e-mail/ > > (è di febbraio, ma non mi sembra di averlo visto passare in lista; mi > scuso qualora fosse successo) > > -- > > -- EN > > https://www.hoepli.it/libro/la-rivoluzione-informatica/9788896069516.html > ====================================================== > Prof. Enrico Nardelli > Presidente di "Informatics Europe" > Direttore del Laboratorio Nazionale "Informatica e Scuola" del CINI > Dipartimento di Matematica - Università di Roma "Tor Vergata" > Via della Ricerca Scientifica snc - 00133 Roma > home page: https://www.mat.uniroma2.it/~nardelli > blog: https://link-and-think.blogspot.it/ > tel: +39 06 7259.4204 fax: +39 06 7259.4699 > mobile: +39 335 590.2331 e-mail: nardelli(a)mat.uniroma2.it <mailto:nardelli@mat.uniroma2.it> > online meeting: https://blue.meet.garr.it/b/enr-y7f-t0q-ont > ======================================================

12 33